Ahoana no ahafahan'ny mpikaroka manamarina, mahita ary mampitaha ny fanitsiana AI

Ny mpanoratra: Enes Altuncu, ea483@kent.ac.uk (University of Kent, UK) Virginia N. L. Franqueira, V.Franqueira@kent.ac.uk (University of Kent, UK) Shujun Li, S.J.Li@kent.ac.uk (University of Kent, UK) Ny mpanoratra: Enes Altuncu, ea483@kent.ac.uk (Ny Oniversiten'i Kent, UK) Virginia N. L. Franqueira, V.Franqueira@kent.ac.uk (University of Kent, Angletera) dia mpianatra ao amin'ny University of Kent. Shujun Li, S.J.Li@kent.ac.uk (Ny Oniversiten'i Kent, UK) Abstract Ny Abstract Ny fandrosoana farany tao amin'ny AI, indrindra ny fampianarana goavana, dia nanampy tamin'ny fampitomboana goavana amin'ny famoronana haino aman-jery fototra vaovao (video, sary, ary feo) sy ny fanodinana ny haino aman-jery misy, izay nitondra ny famoronana ny teny vaovao "deepfake". Miorina amin'ny boky sy ny fitaovana fikarohana amin'ny teny Anglisy sy Shinoa, ity lahatsoratra ity dia manome fanadihadiana feno momba ny deepfake, ahitana lafiny maro manan-danja amin'ity hevitra mitranga ity, anisan'izany ny 1) famaritana samihafa, 2) fampisehoana sy fitsipika ampiasaina matetika, ary 3) daty mifandraika amin'ny deepfake, fanamby, fifaninanana ary tsipika. Anko : Deepfake, Survey, Definition, Datasets, Benchmarks, Challenge, fifaninanana, Standard, Performance Metrics. Ny teny fototra 1 Ny fampidirana Ny fandrosoana farany tao amin'ny AI sy ny fianarana milina dia nanatsara ny fahafahana mamokatra haino aman-jery mahomby kokoa, ohatra, lahatsary, sary, ary feo. Indrindra indrindra, ny fomba fianarana maoderina maoderina no nahatonga ny famoronana "deepfakes", fitaovana voajanahary na synthesized izay tsy mora ho fantatra amin'ny mason'ny olombelona. Na dia zava-mahazatra vao haingana aza (nitranga voalohany tamin'ny faran'ny 2017), ny fitomboany dia tena mahatalanjona. Araka ny tatitra 2019 sy 2020 Deeptrace momba ny toe-javatra ny deepfake [ Ny isan'ny lahatsary deepfake ao amin'ny aterineto miteny anglisy dia niakatra avy amin'ny 7,964 (Dezambra 2018) ho 14,678 (July 2019) ho 85,047 (Dezambra 2020), izay mampitombo 968% avy amin'ny 2018 ka hatramin'ny 2020. 2 Amin'ity asa ity, dia manara-maso ny tontolo iainana fikarohana mifandraika amin'ny deepfake isika amin'ny lafiny isan-karazany, anisan'izany ny metriky sy ny fitsipika, ny tahirin-kevitra, ny fanamby, ny fifaninanana ary ny valin-teny. Ankoatra izany, manome fanadihadiana meta momba ny lahatsoratra fanadihadiana 12 voafidy mifandraika amin'ny deepfake izay ahitana lafiny maro hafa noho ireo voalaza tamin'ny fomba rafitra, toy ny fampitahana ny vokatra, ny fanamby fototra, ary ny toro-hevitra. Na dia tena malaza aza ny teny, tsy misy fifanarahana eo amin'ny famaritana ny "deepfake" ary ny sisintany eo amin'ny deepfakes sy ny tsy-deepfakes dia tsy mazava mazava. Ho an'ity fanadihadiana ity, dia mandray fomba fijery mahazatra kokoa isika mba ahitana ny endrika rehetra amin'ny haino aman-jery voajanahary na synthetic izay heverina ho deepfakes amin'ny lafiny lehibe kokoa. Hahita lohahevitra mifandray akaiky ihany koa isika, anisan'izany ny biometrika sy ny multimedialy forensics, satria matetika ampiasaina ny deepfakes mba hanomboka fanafihana fanehoan-kevitra amin'ny rafitra biometrika mifototra amin'ny fanamarinana sy ny fanamarinana ny deepfakes dia azo hever 1.1 Ny famaritana ny teny Deepfake Araka ny anarany, ny teny "deepfake" dia avy amin'ny fiaraha-miombon'ny "deep" (manondro ny (DL)) sy "fake". Matetika ampiasaina izany mba hifandray amin'ny fanitsiana ny haino aman-jery efa misy (foto, lahatsary sy / na feo) na famokarana haino aman-jery vaovao (synthetic) amin'ny fampiasana ny fomba fampiasana DL. Ny angon-drakitra deepfake matetika dia fanehoan-kevitra diso, fanehoan-kevitra diso, ary lahatsary diso izay mampifandray sary diso sy fanehoan-kevitra diso. Raha manana "fake" ao amin'ny teny dia manondro ny haino aman-jery diso na synthesized, misy fampiharana goavana maro amin'ny teknolojia deepfake, ohatra, ho an'ny fampiroboroboana sy ny zavakanto. Amin'izany fomba izany, ny teny iray hafa Na izany aza, tsy voafehin'ny ankamaroan'ny olona ny teny vaovao. Ny fampianarana lalina 60 Ankoatra ny tsy fisian'ny famaritana ankapobeny, araka ny voalaza etsy ambony, ny sisintany eo amin'ny deepfakes sy ny tsy-deepfakes dia tena tsy mazava. Voalohany, ny fanadihadiana ny deepfakes matetika dia manaraka fomba maro mitovy amin'ny fanadihadiana ny deepfakes nentim-paharazana izay nateraky tsy mampiasa teknika DL. Ny fomba fanadihadiana ambony ihany koa dia nanomboka mampiasa DL mba hanatsarana ny fahombiazan'izy ireo, fa tsy voatery mahalala ny fomba famoronana media tanjona (sy mainty na tsy). Amin'ny lafiny sasany, afaka milaza ianao fa ny fanadihadiana deepfakes dia tsy ahitana ny fampandrosoana fomba manokana amin'ny deepfakes (afa-tsy ny mpikaroka sasany no manapa-kevitra ny manao izany), fa ny detector mahery sy ankapobeny izay afaka mifandray amin'ny media mainty na tsy mainty. Izany dia azo jerena amin' Faharoa, azonao atao ny milaza fa ny teny "deep" ao amin'ny "deepfake" dia tsy voatery mifandray amin'ny fampiasana ny "deep learning", fa na inona na inona "deep" (toy ny rafitra) teknolojia izay mamorona haino aman-jery tena azo itokisana. Nolazain'izy ireo fa ny deepfake dia fanitsiana audio-visual mampiasa "fitaovana ara-teknika ... sy ny teknika". Ary , mifandray amin'ny "fanitsiana amin'ny sehatra ambany amin'ny haino aman-jery voarakitra amin'ny rindrambaiko (na tsy misy rindrambaiko) ahafahana manamaivana, mampihena, mamerina na mampifandray indray ny votoaty". Ity fahatakarana lehibe kokoa momba ny "deepfake" ihany koa no nafenin'ny mpanao lalàna ho an'ny lalàna vaovao mifehy ny deepfakes ratsy. ohatra, ny lalàna roa manaraka ireto Etazonia dia mamaritra "deepfakes" toy izao manaraka izao: 9 Ny fahadisoana Ny fahadisoana Amin'ity taona 2018 ity dia tsy maintsy ampiharina amin'ny Malicious Deep Fake Prohibition Act 1. Araka ny voalaza ao amin'ny § 1041(b) : ” Ny teny hoe "fampihorohoroana goavana" dia midika hoe rakitra audiovisual navoakan'ny fomba izay toa tsy marina ho an'ny mpijery mahasoa ny rakitra ho rakitra marina momba ny tenin'ny olona iray na ny fitondran-tena. Ny lalàmpanorenan'ny Lalàmpanorenan'ny Lalàmpanorenan'ny Lalàmpanorenan'ny Lalàmpanorenan'ny Lalàmpanorenan'ny Lalàmpanorenan'ny Lalàmpanorenana 2019 dia: §1041.(n).(3): “ Ny fehezan-dalàna dia midika hoe fandefasana lahatsary, horonantsary, feo, sary elektronika, na fampisehoana ara-teknika momba ny feo na ny fitondran-tena izay mifototra amin'izany. (A) izay toa mampiseho ny teny na ny fitondran-tena rehetra avy amin'ny olona izay tsy nanaiky ny teny na ny fitondran-tena toy izany; ary ” (B) ny famokarana dia miankina amin'ny fitaovana ara-teknika, fa tsy ny fahaiza-manao ny olona iray hafa mba ho voajanahary na feon'ny olona toy izany. Araka ny hitantsika ao amin'ny famaritana ara-dalàna etsy ambony momba ny "deepfake", ny fampiasana DL toy ny teknolojia dia tsy voalaza loatra. 1.2 Ny vokatra sy ny fandraisana anjara Miorina amin'ny adihevitra etsy ambony momba ny famaritana ny deepfake, dia afaka mahita fa tsy dia tsotra na mahasoa foana ny manavaka deepfakes avy amin'ny tsy-deep fakes. Ankoatra izany, noho ny fifantohana amin'ny fanombanana sy ny fampitahana ny fahombiazan'ny, ny sisintany eo amin'ny deepfakes sy ny tsy-deep fakes dia maivana kokoa. Izany dia noho ny DL dia endrika manokana (mainty kokoa) amin'ny fianarana milina (ML), ary noho izany, ny DL sy ny tsy-deep ML dia mizara hevitra maro mahazatra, metriky ary dingana maro. Na dia eo aza ny zava-misy fa mety ho fantatra amin'ny lafiny mahazatra kokoa, amin'ity asa ity dia manana fifantohana mahazatra kokoa isika mba hisorohana lohahevitra be loatra. Ho an'ny metriky sy ny fitsipika, dia nifidy ny ahitana ireo izay ampiasaina matetika amin'ny fanombanana ny fomba ML ankapobeny sy ireo izay voafaritra manokana amin'ny fanombanana ny fomba famoronana na fametrahana deepfake. Ho an'ny angon-drakitra, fanamby, fifaninanana ary toro-hevitra, dia nanadihady ireo mifandray amin'ny famoahana haino aman-jery mifandray amin'ny fanadihadiana sy ireo miaraka amin'ny fametrahana mazava ny teny "deepfake" na teny mitovy. Ho an'ny meta-review, dia nihevitra fotsiny izahay ny tatitra momba ny fanadihadiana izay niresaka mazava ny teny hoe "deepfakes" ao amin'ny meta data (ny lohateny, abstract sy ny teny fanalahidy). 2 Ny fomba fanao Ny tatitra momba ny fikarohana voarakitra ao amin'ity fanadihadiana ity (ohatra ny tatitra momba ny fikarohana mifandraika amin'ny fanadihadiana) dia hita amin'ny alàlan'ny fikarohana ara-tsiansa ao amin'ny tahirin-kevitra, Scopus sy China Online Journals (COJ) . The following search queries were used to perform the searches on Scopus and COJ, respectively: 3 (deepfake* OR deep-fake* OR “deep fake*”) ary (fanadihadiana na fanadihadiana na fanadihadiana na systemati* na SoK) (deepfake OR 深度伪造) AND (综述 OR 进展) Ny fikarohana dia niverina lahatsoratra 41 amin'ny teny Anglisy sy lahatsoratra 15 amin'ny teny Shinoa. Avy amin'ireto lahatsoratra ireto dia voafidy ho an'ny fanadihadiana dimy voasoratra amin'ny teny Anglisy sy efatra voasoratra amin'ny teny Shinoa. Ny fanamby, ny fifaninanana ary ny toro-hevitra mifandraika amin'ny Deepfake dia voamarina amin'ny alàlan'ny loharanom-baovao maromaro: ny taratasy fanadihadiana voafidy, ny taratasy fikarohana avy amin'ny fanangonana manokana avy amin'ny mpandray anjara, ny fikarohana amin'ny Google Web, ary ny fanaraha-maso an-tserasera amin'ny tranonkala amin'ny fifanarahana lehibe mifandraika amin'ny AI natao tamin'ny 2020 sy 2021 (raha ny fanamby sy ny fifaninanana toy izany dia voatendry matetika). , ICCV, CVPR, AAAI, ICML, ICLR, KDD, SIGIR, WWW, ary maro hafa. Ankoatra izany, ny lisitry ny angon-drakitra dia naorina mifototra amin'ny tatitra fanadihadiana voafidy sy ny fanamby, fifaninanana, ary benchmarks. Ny fitsipika manan-danja dia voatazona indrindra amin'ny alàlan'ny tatitra fikarohana voarakitra amin'ity fanadihadiana ity, ny fahalalana manokana momba ny mpandray anjara, ary ny fikarohana amin'ny Google Web. Ho an'ny fampisehoana metriky, dia ahitana ireo ampiasaina matetika mifototra amin'ny fitsipika manan-danja, ny tatitra fanadihadiana, ary ny fanamby, fifaninanana, ary ben 4 3 Deepfake mifandraika amin'ny fampisehoana sy ny fitsipika Ao amin'ity fanadihadiana ity, dia mifantoka amin'ny fanombanana ny vokatra sy ny mampitaha ny fomba famokarana sy fametrahana deepfake. Ny metriky ampiasaina amin'ny fanombanana ny vokatra toy izany dia ny fototry ny adihevitra. Ao amin'ity fizarana ity, dia manara-maso ny metriky vokatra izay matetika ampiasaina amin'ny fanombanana ny algorithm famokarana sy famantarana deepfake. Tsarovy fa ny metriky rehetra voatahiry ao amin'ity fizarana ity dia matetika ihany koa ampiasaina amin'ny fanombanana ny vokatra amin'ny rafitra mitovy izay tsy ampiasaina amin'ny famokarana na famantarana deepfakes. Noho izany, ity fizarana ity dia azo heverina ho fampianarana fohy momba ny Ao amin'ny andininy farany, dia miresaka fohy ihany koa ny fomba ahitana ny endri-javatra mifandraika amin'ny fitsipika ofisialy. Amin'ny anarana hoe "standards" isika dia midika hoe fitsipika voafaritra araka ny dingana ofisialy, matetika amin'ny alalan'ny vatana iray na maromaro miorina fitsipika, toy ny International Organization for Standardization (ISO). Ny Firenena Mikambana Electrotechnical Commission (IEC) Mariho fa isika dia mihevitra fa maro ireo antontan-taratasy voafaritra ho an'ny fitsipika avy amin'ny fikambanana fitsipika, ohatra, ny International Telecommunication Union (ITU) ny fanapahan-kevitra sy ny tatitra ara-teknika ISO (TRs). 5 6 7 3.1 Ny Matrix Mifanohitra Deepfake detection dia voalohany indrindra ny olana amin'ny famaritana binary. Ny famaritana binary dia mandray fidirana izay Ary ary avy amin'ny vidin'ny binary izay manondro izany ho Ary Ohatra, ny rafitra fanadihadiana deepfake dia handray sary ahiahy ho toy ny fidirana izay mety ho Ary Ny vokatra Ary . Raha ny marina dia tsara Ny tena ratsy Ny fanehoan-kevitra tsara Ny fanehoan-kevitra ratsy Ny tena marina Ny tena marina Ny fanehoan-kevitra ho an'ny false Ny tena zava-misy Ny fitaovana fototra ampiasaina amin'ny fanombanana ny famaritana binary dia ny Izany dia manazava ny fahombiazana sy ny tsy fahombiazana amin'ny modely famaritana. eo amin'ny andaniny sy eo amin'ny andaniny dia ny roa Ny valiny dia ny famaritana (Ny tena tsara sy ny tena ratsy) rehefa mifanaraka amin'ny tena sy ny voalaza (Fahadisoana tsara sy fahadisoana ratsy) raha tsy mifanaraka amin'ny zava-misy sy ny voalaza. mampiseho ny matric confusion ho an'ny binary deepfake classifier (detector). Ny sela roa amin'ny maitso, TP (ny isan'ny ) ary TN ( ny isan'ny ), manondro ny valin'ny famaritana, ary ny sela roa amin'ny mena, FN (ny isan'ny ) ary ny FP ( ny isan'ny ), mampiseho karazana fahadisoana roa samihafa amin'ny fametrahana vokatra tsy marina. confusion matrix Amin'izao fotoana izao Ny fikasana Ny marina / ny marina / ny marina Tsy marina / tsy marina / tsy marina 1 true positives true negatives false negatives false positives Table 1: Matrix Confusion ho an'ny binary classifier ho an'ny fanadihadiana deepfake. fake (predicted) real (predicted) fake (actual) TP FN real (actual) FP TN Ny zava-misy dia Ny TP Ny Firenena Mikambana Ny tena zava-misy Ny FP Ny Tompo 3.2 Ny fanamafisana sy ny famerenana Miorina amin'ny fitsipika efatra fototra voalaza ao amin'ny fizarana , izany hoe, TP, TN, FP sy FN, dia mamaritra ny fampisehoana roa manan-danja ho an'ny famaritana binary - Ary . 3.1 precision recall Ny fahamarinana amin'ny famaritana binary dia voafaritra ho toy ny ampahany amin'ny Samples eo amin'ny rehetra ny Amin'ny ankapobeny, ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ( Ny Raha ny marina dia tsara Ny fanehoan-kevitra tsara 1 Rehefa tsy mitovy amin'ny fanandramana ny tahan'ny "tsy voajanahary" eo amin'ny valin'ny tsara sy ny tsy voajanahary, matetika dia mahasoa ny fanitsiana ny lanjany amin'ny lainga, izay mitarika amin'ny Amin'ny alàlan'ny fampiasana ny fampiasana ny fanafody ( Aiza ny 0 dia lanjany voafaritra amin'ny fifandraisana eo amin'ny sampana ratsy sy ratsy. weighted precision 2 Ny Ny famerenana ny famaritana binary dia ny ampahany amin'ny Ny sampana eo amin'ny samples, as shown in Eq. ( Ao amin'ny matric confusion, dia ny ampahany amin'ny sampana marina ao amin'ny andininy voalohany. Ny fanehoan-kevitra tsara Raha ny marina dia tsara 3 Asehoy ny ohatra amin'ny famaritana binary izay manambara raha ny sary avy amin'ny tahirin-kevitra izay ahitana sary mainty sy tena (heverina) dia diso na tsia. Ny fahamarinana ny famaritana dia ny ampahany amin'ny sary voasoratra tsara eo amin'ny sary rehetra voasoratra ho mainty. Amin'ny lafiny iray, ny famerenana dia ny ampahany amin'ny sary mainty voasoratra ao amin'ny tahirin-kevitra. 3.3 Ny vidin'ny tena sy ny tsy marina Mifantoka amin'ny fanehoan-kevitra tsara azo antoka, dia azontsika atao ihany koa ny mamaritra marika roa: (TPR) koa ny antsoina hoe (CDR), raha oharina amin'ny ampahany amin'ny fanehoan-kevitra tsara eo amin'ny fanehoan-kevitra tena tsara sy (FPR) koa ny antsoina hoe (FAR), toy ny ampahany amin'ny ohatra tsara voalaza eo amin'ny ohatra ratsy marina, araka ny hita ao amin'ny Eqs. ( Ary true positive rate correct detection rate false positive rate false alarm rate 4 ( ). In the confusion matrix, TPR is the fraction of predicted positive samples in the first row and FPR is the fraction of predicted positive samples in the second row. Note that TPR is basically a different name for (Amin'ny ankapobeny) Ary ny 5 recall 3 3.4 True and False Negative Rates Toy izany koa amin'ny tahan'ny tena tsara sy ny tsy marina tsara, dia afaka mamaritra ny tahan'ny roa hafa izay mifantoka amin'ny vokatra ratsy azo antoka: (TNR) izay mampiseho ny ampahany amin'ny sampana ratsy voalaza eo amin'ny sampana ratsy marina, ary (FNR) izay mampiseho ny ampahany amin'ny fanambaran'ny fihenam-bidy ratsy eo amin'ny fanambaran'ny tena tsara, araka ny voalaza ao amin'ny Eqs. ( Ary ny ( Ny true negative rate false negative rate 6 7 3.5 Ny fahatsapana sy ny fahatsapana manokana Ao amin'ny fampiharana sasany amin'ny famaritana binary, indrindra ao amin'ny biolojia sy ny fanafody, ny TPR sy ny TNR dia ampiasaina matetika, ary matetika izy ireo dia antsoina hoe Ny TPR ary (TNR) Ny fifantohana amin'ireo teny roa ireo dia amin'ny karazana fahamarinana roa amin'ny vokatra voalaza. Ireo dia tsy ampiasaina amin'ny fikarohana mifandraika amin'ny fako, noho izany dia tsy hitarika azy ireo amin'ny sisa amin'ity lahatsoratra ity. sensitivity specificity 3.6 Ny tahan'ny fahadisoana mitovy Ny fifantohana amin'ny tahan'ny fahadisoana dia midika fa mila mandinika ny FPR sy ny FNR. Ireo tahan'ny roa ireo dia matetika mifanohitra amin'ny hafa mba hampihenana ny tahan'ny iray amin'ny ankapobeny dia miteraka fitomboana amin'ny hafa. Noho izany, fa tsy miezaka ny hampihenana ny tahan'ny fahadisoana roa amin'ny fotoana iray, izay tsy azo atao amin'ny ankapobeny, ny andraikitry ny zava-misy kokoa amin'ny fampiharana ara-toekarena dia ny mahita ny fitoviana tsara mba ho eo ambany ny lalana azo ekena. Ao amin'ny fampiharana sasany, toy ny biometrika, ny olona dia tena liana amin'ny fametrahana ny antsoina hoe Na dia Ny EER / CER dia tsy voatery ho marika tsara ho an'ny fampiharana sasany, indrindra rehefa misy karazana fahadisoana roa samihafa amin'ny lanjany, ohatra, amin'ny fanandramana deepfakes manan-danja (ohatra vaovao diso izay mety hisy fiantraikany amin'ny fomba fandefasana ny fangatahany ny olona) matetika isika dia afaka mandresy bebe kokoa ny false positive (false alarms) noho ny false negative (missed alarms). equal error rate crossover error rate 3.7 Ny fahamarinana sy ny F-Score Ankoatra ny EER / CER, misy ihany koa ny metriky hafa izay miezaka mamaritra ireo karazana fahadisoana roa ireo, mba hanome fampisehoana tsara kokoa momba ny fampisehoana ankapobeny amin'ny famaritana binary. Ary Ny antsoina koa Ny metrika roa dia azo voafaritra mifototra amin'ny isa efatra fototra (TP, TN, FP, ary FN). accuracy F-score F-measure Ny fahamarinana amin'ny famaritana binary dia voafaritra ho toy ny ampahany amin'ny Ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ( Ny Ny fanantenana marina 8 Ny F-poizina amin'ny famaritana binary dia tena fianakaviana metriky. Ny endriny ankapobeny dia azo lazaina mifototra amin'ny parameter Araka ny voalazan'ny tsiambaratelo ( ). β 9 Ny famoahana ampiasaina amin'ny ankapobeny indrindra amin'ny vokatra F dia ny antsoina hoe , izay amin'ny ankapobeny ny F-score amin'ny Amin'ny ankapobeny, araka ny voalazan'ny tsiambaratelo ao amin'ny Eq. ( Ny F1-score β 10 3.8 Receiver Operating Characteristic Curve sy ny faritra eo ambanin'ny curve (ROC) curves dia matetika ampiasaina mba handrefesana ny fampisehoana ny binary classifiers izay mamoaka ny isa (na mety) ny faminaniana. Receiver operating characteristic Jereo ny manaraka ireto: Aoka be the set of all test samples and let the output scores Ny ) (for all ∈ ) mandainga ao amin'ny interval [ Ao amin'ny lafiny tena izy. Izany dia ] be a prediction threshold for the model, and assume that the classifiers works as follows for all Ny : S f s s S a, b t a, b s S It is easy to see that, for = , all the samples will be classified as positive, leading to FN = TN = 0 so TPR = FPR = 1; while for = , all the samples will be classified as negative, leading to FP = TP = 0 so TPR = FPR = 0. For other threshold values between and , the values of TPR and FPR will normally be between 0 and 1. By changing from to continuously, we can normally get a continuous curve that describes how the TPR and FPR values change from (0,0) to (1,1) on the 2D plane. This curve is the ROC curve of the binary classifier. t a t b a b t a b For a random classifier, assuming that ( ) distributes uniformly on [ ] for the test set, we can mathematically derive its ROC curve being the TPR = FPR line, whose area under the ROC curve (AUC) is 0.5. For a binary classifier that performs better than a random predictor, we can also mathematically prove that its AUC is always higher than 0.5, with 1 being the best possible value. Note that no binary classifier can have an AUC below 0.5, since one can simply flip the prediction result to get a better predictor with an AUC of 1 − AUC. The relationship between the ROC and the AUC is graphically illustrated in Figure . f s a, b 1 3.9 Log Loss Ny metrik'asa hafa ampiasain'ny fampisehoana ho an'ny famaritana binary izay afaka miverina amin'ny valin'ny mety ho an'ny marika voalaza dia . For a binary classification with a true label ∈ {0*,* 1} and an estimated probability = Pr( = 1), the log loss per sample is the negative log-likelihood of the classifier given the true label, defined as shown in Eq. ( ). log loss y p y 12 Given a testing set with samples, the log loss score of a binary classifier can be calculated using Eq. ( ), where is 1 if the -th sample is true and 0 if false, and ˆ Noho izany, ny mety ho azo antoka = 1. n 13 yi i y i yi 3.10 Extension to Multi-class Classifiers All metrics that are defined based on the four basic values TP, TN, FP and FN can be easily extended to by considering the prediction to be true or false individually with respect to each class. For example, if the system is classifying animals (cats, dogs, horses, lions, tigers, etc.), then a true positive prediction of an image to be of a cat, would simultaneously be true negative predictions for the remaining classes (dogs, horses, lions, tigers, etc.). If an image of a cat is incorrectly predicted to be that of a dog, it would be a false negative with respect to a cat, a false positive with respect to a dog, and a true negative with respect to all other classes. multi-class classification 3.11 Perceptual Quality Assessment (PQA) Metrics By definition, the main goal of deepfakes is to make it hard or impossible for human consumers (listeners or viewers) to distinguish fake media from real media. Therefore, when evaluating the quality of deepfake media, the quality perceived by human consumers of the media is key. This calls for subjective assessment of the perceptual quality of the deepfake media as the “gold standard”. The most widely used subjective perceptual quality assessment (PQA) metric for audio-visual signals is (MOS), izay ampiasain'ny vondrom-piarahamonina amin'ny fanodinana famantarana sy fifandraisana multimédia, anisan'izany ny fahitalavitra nomerika sy ny fampiharana mpanjifa hafa mifandraika amin'ny multimédia. Araka ny anarany, ny MOS dia voafaritra amin'ny alalan'ny fanombanana ny valin'ny olona maro, matetika manaraka ny habetsaky ny 1 ka hatramin'ny 5 na eo anelanelan'ny 0 ka hatramin'ny 100. ) and also for evaluating and comparing the quality (realness/naturalness) of deepfake datasets (see Section ). mean opinion score 5.2 4.6 Amin'ny maha-zava-dehibe ny PQA, ny MOS dia voafaritra amin'ny ITU . There are also ITU standards defining more specific subjective Video Quality Assessment (VQA) metrics and the standard procedures one should follow to conduct VQA user studies, e.g., ITU-T Recommendation P.910 “Subjective video quality assessment methods for multimedia applications” . Note that the ITU standards focus more on traditional perceptual quality, i.e., how good a signal looks or sounds, even if it looks or sounds not real (e.g., too smooth). On the other hand, for deepfakes, the focus is rather different because what matters is the realness and naturalness of the created media, i.e., how real and natural it looks or sounds, even if it is of low quality. To some extent, we can also consider realness and naturalness as a special aspect of perceptual quality. 8 9 One major problem of subjective PQA metrics like MOS is the need to recruit human judges and to have a well-controlled physical testing environment and protocol, which are not easy for many applications. To help reduce the efforts and costs of conducting PQA-related user studies, various objective PQA metrics have been proposed, where the term “objective” refers to the fact that such metrics are human-free, i.e., automatically calculated following a computational algorithm or process. Depending on whether a reference exists, such objective PQA metrics can be largely split into three categories: full-reference (FR) metrics (when the original “perfect-quality” signal is available as the reference), reduced-reference (RR) metrics (when some features of the original “perfect-quality” signal are available as the reference), and no-reference (NR) metrics (when the original signal is unavailable or such an original signal does not exist). For deepfakes, normally NR or RR metrics are more meaningful because the “fake” part of the word means that part of the whole data does not exist in the real world, hence a full reference cannot be obtained. RR metrics are still relevant because deepfakes are often produced for a target’s specific attributes (e.g., face and voice), where the reduced reference will be such attributes. NR metrics will be useful to estimate the realness and naturalness of a deepfake, simulating how a human judge would rate it in a controlled subjective PQA user study. PQA is a very active research area and many PQA metrics have been proposed, some of which have been widely used in real-world products and services, e.g., (MSE), (PSNR) and (SSIM) for FR PQA of digitalimages and videos defined as in Eqs. (14), (15), and (16), respectively, where X = {xi} n i is the reference (the original signal), Y = {yi} n i is the signal whose visual quality is assessed, n is the number of pixels in X and Y , L is the maximum possible pixel value of X and Y (e.g., 255 for 8-bit gray-scale images), c1 = (k1L) 2 and c2 = (k2L) 2 ) are two stabilising parameters (k1 = 0.01 and k2 = 0.03 by default). For more about PQA metrics for different types of multimedia signals, we refer readers to some relevant surveys [3, 51, 72]. mean squared error peak signal-to-noise ratio structural similarity index measure 3.12 More about Standards Many of the basic performance metrics described in this section have been widely used by deepfake researchers as de facto standards, e.g., EER, log loss and MOS have been widely used in deepfake-related challenges (see Section ). Also, the combination of precision, recall and F1-score has been widely used to assess performance of binary classifiers. While there have been a number of ITU standards on PQA to date, there does not seem to be many standardisation efforts on the performance metrics for evaluation of binary classifiers. This was the case until at least 2017, when ISO and IEC jointly set up the ISO/IEC JTC 1/SC 42 , a standardisation subcommittee (SC) focusing on AI under ISO/IEC JTC 1 , ny komity ara-teknika mifandraika amin'ny fanamafisana ny "technology fampahalalam-baovao". 5 10 11 One recent effort that ISO/IEC JTC 1/SC 42 made is to produce the ISO/IEC TR 24029-1:2021 “Artificial Intelligence (AI) – Assessment of the robustness of neural networks – Part 1: Overview” , tatitra ara-teknika (TR) izay mandrakariva ahitana hevitra, fomba sy metrika maro ampiasaina amin'ny fanombanana ny fahombiazan'ny fampisehoana. Na dia misy ny "network neural" ao amin'ny lohateny aza ny tatitra ara-teknika, ny ankamaroan'ny toro-hevitra, fomba sy metrika momba ny fanombanana ny fahombiazan'ny fampisehoana dia mahazatra amin'ny modely rehetra mifehy ny fianarana amin'ny milina. 12 Raha ny momba ny metrik'ny fampisehoana, ny zavatra roa hafa ao amin'ny ISO/IEC JTC 1/SC 42 izay mendrika ny fanehoan-kevitra dia toy izao manaraka izao: ISO/IEC DTS (Draft Technical Specification) 4213 "Information technology - Artificial intelligence - Assessment of machine learning classification performance" (Ny teknolojia fampahalalam-baovao - Intelligence Artificial - Fanombanana ny fahombiazan'ny famaritana ny fianarana milina) 13 ISO/IEC AWI (Approved Work Item) TS (Technical Specifications) 5471 “Artificial intelligence – Quality evaluation guidelines for AI systems” 14 While the ISO/IEC JTC 1/SC 42 was created very recently, another standardisation subcommittee under ISO/IEC JTC1 has a much longer history of nearly 20 years: the ISO/IEC JTC 1/SC 37 that focuses on biometrics-related technology. This standardisation subcommittee is highly relevant for deepfake since deepfake faces can be used to spoof biometrics-based user authentication systems. In this context, the following three standards are of particular relevance: 15 : This standard covers general metrics about evaluating biometric systems. Two major metrics in this context are (FAR) and (FRR), which refer to the standard FPR and FNR, respectively. This standard also deprecates the use of single-number metrics including the EER and AUC (which were widely used in biometrics-related research in the past). ISO/IEC 19795-1:2021 “Information technology – Biometric performance testing and reporting – Part 1: Principles and framework” 16 false accept rate false reject rate : This standard defines a general framework about (PAD) mechanisms, where the term “ ” refers to the “ ”. It focuses on biometric recognition systems, where a PAD mechanism is a binary classifier trying to predict presentation attacks (also called attack presentations, e.g., fake faces) as positive and bona fide (real) presentations as negative. ISO/IEC 30107-1:2016 “Information technology – Biometric presentation attack detec-tion – Part 1: Framework” 17 presentation attack detection presentation attack presentation of an artefact or of human characteristics to a biometric capture subsystem in a fashion intended to in-terfere with system policy : This standard defines a number of special performance metrics for evaluating PAD mechanisms standardised in the ISO/IEC 30107-1:2016. Three such metrics look at error rates: (APCER) referring to the standard FPR, (NPCER/BPCER) referring to the standard FNR, and (ACER) that is defined as the average of the APCER and the NPCER/BPCER. Such metrics have been used in biometrics-related challenges such as Face Anti-spoofing (Presentation Attack Detection) Challenges . When deepfake images or videos are used to spoof a biometric system, such standardised metrics will become relevant. ISO/IEC 30107-3:2017 “Information technology – Biometric presentation attack detection – Part 3: Testing and reporting” 18 attack presentation classification error rate normal/bona fide presentation classification error rate average classification error rate 19 3.13 Discussion: Performance Metrics & Standards This section provided a comprehensive summary of performance metrics used for evaluating and bench-marking binary classifiers. It is rare that all such metrics are used for a specific application. Instead, one or several are chosen based on specific needs. For a deepfake detection system as a binary classifier, many researchers have chosen to use overall metrics such as accuracy, AUC, EER and log loss, but the combination of precision, recall and F1-score is also common. Some deepfake-related challenges and competitions have introduced their own specific metrics, some of which will be described in Section . The use of different performance metrics can make comparison of different reported results more difficult, so we hope the expected new ISO/IEC standard particularly ISO/IEC 4213 will help. 5 It is worth mentioning that, in addition to evaluating performance of deepfake detectors, the introduced performance metrics for evaluating binary classifiers can also be used to evaluate performance of deepfake generation methods by considering how deepfake detectors fail. For instance, organisers of the Voice Conversion Challenge 2018 and 2020 used this approach to benchmark how well voice conversion (VC) systems can generate high-quality fake speech samples. Another point we would like to mention is that for deepfake videos there are two levels of performance metrics: those at the frame level (metrics of each frame), and those at the video level (metrics for the whole video). Generally speaking, the latter can be obtained by averaging the former for all frames, potentially following an adaptive weighting scheme, so that more important (key) frames will be counted more. 4 Ny angon-drakitra mifandraika amin'ny fikarohana In this section, we cover all deepfake-related datasets we identified from the meta-review of deepfake-related survey papers, deepfake-related challenges, competitions and benchmarks covered, one online collection of deepfake-related datasets on GitHub , and the co-authors’ personal collections. Table shows basic information about these datasets. We explain them in four categories: deepfake image datasets, deepfake video datasets, deepfake audio/speech datasets, and hybrid deepfake datasets (mainly mixed image and video datasets). 20 2 Note that many datasets of real (authentic) media were also used by deepfake researchers for two purposes. First, any detectors would need both fake and real media to demonstrate their performance. Second, real media have also been used to train deepfake generators as the training set. In this section, we include only datasets containing deepfake media, some of which contain both deepfake and real media. Some datasets, especially those created for deepfake-related challenges and competitions, have separate subsets for training and evaluation (testing) purposes. The split is necessary for such challenges and competitions, but not very useful for people who just want to use such datasets. Therefore, in this section when introducing such datasets we will ignore that level of details and focus on the total number of data including the number of real and fake samples. 4.1 Deepfake Image Datasets [ ]: This dataset contains 4,310 images, including 2,300 real images and 2,010 fake images created using FaceSwap and the SwapMe iOS app (now discontinued). SwapMe and FaceSwap dataset 78 21 [ ]: This dataset contains 131,500 face images, including 78,500 images extracted from 150 videos in the FaceForensics dataset and 53,000 images extracted from 150 fake videos collected from YouTube. Fake Faces in the Wild (FFW) dataset 32 : This is a number of commercial datasets provided by the Generated Media, Inc., with up to nearly 2.7 million synthetic face images generated by StyleGAN. A free edition with 10,000 128x128 synthetic images is made available for academic research. The website also provides an interactive face generator and an API . The generated.photos datasets have a good diversity: five age groups (infants, children, youth, adults, middle-aged), two genders (male and female), four ethnicities (white, black, Latino, Asian), four eye colours (brown, grey, blue, green), four hair colours (brown, black, blond, gray), three hair length (short, medium, long), facial expressions, three head poses (front facing, left facing, right facing), two emotions (joy and neutral), two face styles (natural, beautified). (According to a number of research papers we read, an earlier 100K-Faces dataset was released by generated.photos for academic research in 2018, which was used by many researchers. This dataset is not currently available any longer.) generated.photos datasets 22 23 24 [ ]: This dataset includes 19,457 face images, including 7,948 deepfake images generated from on 175 forged videos collected online and 11,509 real face images collected from various online sources. (Table 2 of the paper shows the dataset size is 19,509, but the dataset downloaded from pCloud contains just 19,457 images.) MesoNet Deepfake Dataset 1 [ ]: Ity angon-drakitra ity dia ahitana sary an-tsary, efitrano, fiara sy saka 100,000 noforonina avy amin'ny generator GAN nianatra mifototra amin'ny sary tena izy ao amin'ny FFHQ and LSUN datasets ( telo karazana zavatra - efitrano, fiara sy saka - ho an'ny farany). Tsarovy fa ny anaran'ny "100K-Generated-Images" dia tsy tena toy ny mpanoratra [ ] just used this to name a sub-folder of their Google Drive shared space, but it was used in one of the survey papers [ ]. 100K-Generated-Images 30 25 26 30 65 [ ]: This dataset contains 420,053 images of celebrities, including 156,930 real ones downloaded using Google Image API and 263,123 fake face-swapped ones created using two different methods (Nirkin’s method and Auto-Encoder-GAN) Ding et al.’s swapped face dataset 17 Ny ]: This dataset includes 87,000 224x224 face images, generated by processing some StyleGAN-generated synthetic images using the GAN-fingerprint Removal approach (GANprintR) proposed by . It is the replaced version of the dataset, which contains 150,000 face images generated using an earlier version of GANprintR. iFakeFaceDB 48 Neves et al. FSRemovalDB [ ]: Ity dataset ity dia ahitana sary 40,000, antsasaky ny tena sy antsasaky ny tsy marina. Ny sary dia nangonina avy amin'ny loharanom-baovao efatra: ny dataset CelebA-HQ , the Flickr-Faces-HQ dataset , the 100K-Faces dataset (not available any longer, see the description of generated.photos datasets), and . Faces-HQ 21 27 28 29 thisperson-doesnotexist.com [ ]: Ity dataset ity dia ahitana sary 625,537 avy amin'ny famantarana 10,177, miaraka amin'ny toetra 43 manan-karena amin'ny endrika, ny fanazavana, ny tontolo iainana ary ny karazan-tsary. . The 43 attributes include 40 for real images, covering all facial components and accessories (e.g., skin, nose, eyes, eyebrows, lip, hair, hat, eyeglass), and 3 for fake images, covering spoof types, environments and illumination conditions. CelebA-Spoof 75 30 [ ]: This dataset contains 299,039 images, including 58,703 real images sampled from three datasets (FFHQ , CelebA and FaceForensics++ ) and 240,336 fake ones in four main facial manipulation types (identity swap, expression swap, attribute manipulation, and entire synthesis). The images cover two genders (male and female), a wide age groups (the majority between 21 and 50 years old), and both low- and high-quality levels. Diverse Fake Face Dataset (DFFD) 11 31 32 33 4.2 Deepfake Video Datasets [ ]: This dataset contains 620 deepfake face videos, generated by face swapping without manipulation of audio, covering 32 subjects and two quality levels (high and low). DeepfakeTIMIT 35 (FF) [ ]: This dataset contains 1,004 face videos with over 500,000 frames, covering various quality levels and two types of facial manipulation. This dataset is now replaced by the larger FaceForensics++ dataset (see below). FaceForensics 55 (FF++) [ ]: This dataset contains 5,000 face videos with over 1.8 million manipulated frames, including 1,000 real videos (with 509,914 frames) downloaded from YouTube, and 4,000 fake videos created using four face manipulation methods (Deepfakes, Face2Face, FaceSwap and NeuralTextures). The videos cover two genders (male and female), and three quality levels (VGA/480p, HD/720p, and FHD/1080p). FaceForensics++ 56 [ ]: This dataset contains 98 face videos, half (49) are real ones downloaded from Youtube, and the other half are fake ones generated using the FakeApp mobile application (which is now discontinued). The video dataset was created to used to demonstrate a deepfake video detection method based on detection of eye blinking behaviours, so all videos contain at least one eye-blinking event. All fake videos were created by swapping the original face in each of the real videos with the face of the actor Nicolas Cage , thus, only one subject is represented. UADFV dataset 39 34 [ ]: This dataset contains 142 “in the wild” deepfake portrait videos, collected from a range of online sources including news articles, online forums, mobile apps, and research presentations. The videos are diverse, covering the source generative model, resolution, compression, illumination, aspect-ratio, frame rate, motion, pose, cosmetics, occlusion, content, and context. Deep Fakes Dataset 10 [ ]: This dataset contains 5,244 face videos of 66 subjects with both face and voice manipulation. It was released as a preview of the full dataset of the 2020 Deepfake Detection Challenge (DFDC, see below). DFDC (Deepfake Detection Challenge) preview dataset 18 : This dataset contains 1,203 face videos of celebrities, including 408 real videos collected from YouTube with subjects of different ages, ethic groups and genders, and 795 deepfake videos synthesised from these real videos. Celeb-DF v1 35 [ ]: This dataset contains 6,229 face videos of celebrities, including 590 real videos collected from YouTube with subjects of different ages, ethic groups and genders, and 5,639 deepfake videos synthesised from these real videos. Celeb-DF v2 40 [ ]: Ity dataset ity dia ahitana 3 363 lahatsary amin'ny endriky ny endriky ny endriky ny endriky ny endriky ny endriky ny endriky ny endriky ny endriky ny endriky ny endriky ny endriky ny endriky ny endriky ny endriky ny endriky ny endriky ny endriky ny endrika. and JigSaw . DeepFake Detection (DFD) Dataset 20 36 37 [ ]: This dataset contains 60,000 indoor face videos (with 17.6 million frames) generated by face swapping, covering 100 subjects, four skin tones (white, black, yellow, brown), two gen-ders (male and female), different age groups (20-45), 26 nationalities, 7 different angles, 8 face expressions, and different head poses. DeeperForensics-1.0 27 [ ]: This dataset contains 128,154 face videos of 960 subjects, including 23,654 real videos from 3,426 paid actors and 104,500 deepfake videos created using eight different methods (DF-128, DF-256, MM/NN face swap, NTH, FSGAN, StyleGAN, refinement, and audio swap). DFDC (Deepfake Detection Challenge) full dataset 18 10 [ ]: This dataset contains 10,000 high-quality forgery videos, with video- and face-level annotations. The dataset focuses on a more challenging case for forgery detection: each video involves one to 15 individuals, but only some (a minority of) faces are manipulated. FFIW K (Face Forensics in the Wild) dataset 79 FFIW Ny ]: This dataset contains 37,942 videos of paid subjects (395 Koreans and 8 Southeastern Asians), including 62,166 real videos and 175,776 fake ones created using six methods – FaceSwap, DeepFaceLab, FSGAN, First Order Motion Model (FOMM), Audio-driven Talking Face HeadPose (ATFHP) and Wav2Lip. The videos cover a balanced gender ratio and a wide range of age groups. Korean DeepFake Detection Dataset (KoDF) 36 [ ]: This dataset contains 1,737 videos with 1,666,816 frames, including 1,339,843 real frames and 326,973 fake frames generated using the Deep Video Portraits (DVP) [ ] method. The original videos were obtained from three sources: the dataset used in [ ], the Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) [ ], and YouTube. Most videos have a resolution of 1280×720. VideoForensicsHQ 23 34 33 42 [ ]: This dataset contains 7,314 face sequences extracted from 707 deepfake videos that were collected completely from the Internet. It covers diverse scenes, multiple persons in each scene and rich facial expressions. Different from other deepfake video datasets, WildDeepfake contains only face sequences not the full videos. This makes the dataset more like between an image dataset and a video one. We decided to keep it in the video category since the selection process was still more video-focused. WildDeepfake 81 4.3 Deepfake Audio/Speech Datasets Voice conversion (VC) is a technology that can be used to modify an audio and speech sample so that it appears as if spoken by a different (target) person than the original (source) speaker. Obviously, it can be used to generate deepfake audio/speech samples. The biennial Voice Conversion Challenge that started in 2016 is a major challenge series on VC. Datasets released from this challenge series are very different from other deepfake datasets: the deepfake data is not included in the original dataset created by the organisers of each challenge, but in the participant submissions (which are retargeted/fake utterances produced by VC systems built by participants). The challenge datasets also include the evaluation (listening-based) results of all submissions. Some fake utterances may be produced by DL-based VC systems, so we consider all datasets from this challenge series relevant for our purpose of this survey. 38 [ ]: The original dataset created by the challenge organisers was derived from the DAPS (Device and Produced Speech) Dataset [ ]. It contains 216 utterances (162 for training and 54 for testing) per speaker from 10 speakers. Participating teams (17) developed their own VC systems for all 25 source-target speaker pairs, and then submitted generated utterances for evaluation. At least six participating teams used DL-related techniques (LSTM, DNN) in their VC systems (see Table 2 of the result analysis paper ), so the submitted utterances can certainly be considered deepfakes. Voice Conversion Challenge 2016 database 62 47 39 Ny ]: The original dataset created by the challenge organisers was also based on the DAPS dataset. It contains 116 utterances (81 for training and 35 for testing) per speaker from 12 speakers in two different tasks (called Hub and Spoke). Participating teams (23 in total, all for Hub and 11 for Spoke) developed their own VC systems for all 16 source-target speaker pairs, and then submitted generated utterances for evaluation. Comparing with the 2016 challenge, more participating teams used DL-related techniques (e.g., WaveNet, LSTM, DNN, CycleGAN, DRM – deep relational models, and ARBM – adaptive restricted Boltzmann machines) in their VC systems. Voice Conversion Challenge 2018 database 44 [ ]: Ity dataset ity dia mifototra amin'ny dataset Effective Multilingual Interaction in Mobile Environments (EMIME) , a bilingual (Finnish/English, German/English, and Mandarin/English) database. It contains 145 utterances (120 for training and 25 for testing) per speaker from 14 speakers for two different tasks (with 4 × 4 and 4 × 6 source-target speaker pairs, respectively). Participating teams (33 in total, out of which 31 for Task 1 and 28 for Task 2) developed their own VC systems for all source-target speaker pairs, and then submitted generated utterances for evaluation. Comparing with the 2018 challenge, DL-based VC systems were overwhelmingly used by almost all participating teams (WaveNet and WaveGAN among the most used DL-based building blocks). Voice Conversion Challenge 2020 database 70 40 A major set of deepfake speech datasets were created for the (Automatic Speaker Verification Spoofing and Countermeasures) Challenge (2015-2021, held biannually). The datasets for the 2019 and 2021 contain speech data that can be considered deepfakes. ASVspoof 41 [ ]: This dataset is based on the Voice Cloning Toolkit (VCTK) corpus , a multi-speaker English speech database captured from 107 speakers (46 males and 61 females). Two attack scenarios were considered: logical access (LA) involving spoofed (synthetic or converted) speech, and physical access (PA) involving replay attacks of previously recorded bona fide recordings). For our purpose in this survey, the LA scenario is more relevant. The LA part of the dataset includes 12,483 bona fide (real) utterances and 108,978 spoofed utterances. Some of the spoofed speech data for the LA scenario were produced using a generative model involving DL-based techniques such as long short-term memory (LSTM) , WaveNet [ ], WaveRNN [ ], WaveCycleGAN2 [ ]. Note that the challenge organisers did not use the term “deepfake” explicitly, despite the fact that the DL-generated spoofed speech data can be considered as deepfakes. ASVspoof 2019 Challenge database 67 42 43 50 28 58 [ ]: This dataset contains bona fide and spoofed speech data for the logical access (LA) task. The challenge is still ongoing and we did not find a detailed paper on the dataset, so cannot include more details other than its size (7.8 GB after compression). Although we did not see details of the generative algorithms used to produce spoofed speech data, we believe similar DL-based algorithms were used like for the 2019 challenge. ASVspoof 2021 Challenge – Logical Access Database 14 [ ]: In 2021, the challenge included an explicitly defined track on deepfake, but the task description suggests that the organisers of the challenge considered a broader definition of the term “deepfake” by looking at spoofing human listeners rather than ASV (Automatic Speaker Verification) systems. The size of the dataset is 34.5 GB after compression. ASVspoof 2021 Challenge – Speech Deepfake Database 15 Possibly because of the long history and wide participation of the community in the ASVspoof challenges for creating the dedicated datasets, there are very few other deepfake audio/speech datasets. One such dataset was created by a group of researcher from Baidu Research [ ]. This dataset was created to demonstrate a proposed voice cloning method. It is relatively small, and contains 134 utterances, including 10 real ones, 120 cloned ones, and 4 manipulated ones. Another dataset was created by Google AI and Google News Initiative , but it was made part of the ASVspoof 2019 dataset. This dataset contains thousands of phrases spoken by 68 synthetic “voices” covering a variety of regional accents. 5 44 4.4 Hybrid Deepfake Datasets : These datasets were created by the DARPA Media Forensics (MediFor) Program for the 2020 OpenMFC . There are two GAN-generated deepfake datasets, one with more than 1,000 deepfake images and the other with over 100 deepfake videos. The datasets were made available to registered participants of the competition only. NIST OpenMFC (Open Media Forensics Challenge) Datasets 45 46 47 [ ]: This dataset is named as “a versatile benchmark for comprehensive forgery analysis”. It contains 2,896,062 images and 221,247 videos, including 1,457,861 fake images and 121,617 fake videos. The videos and images cover seven image-level and eight video-level manipulation approaches, 36 different types of perturbations and more mixed perturbations, and a large number of annotation labels (6.3 million classification labels, 2.9 million manipulated area annotations and 221,247 temporal forgery segment labels). The dataset is being used for supporting the Face Forgery Analysis Challenge 2021 Ao amin'ny SenseHuman 2021 (3rd Workshop on Sensing, Understanding and Synthesizing Humans) , co-located at the ICCV 2021 conference . ForgeryNet 25 48 49 50 4.5 A Deepfake Dataset Generator [ ]: This is not actually a dataset per se, but a system for producing large datasets more automatically, including generating deepfake datasets. One may argue the automatically generated datasets are fake since they are not produced from real-world scenes. DatasetGAN 74 4.6 Subjective Quality of Deepfakes in Different Databases As mentioned in Section , subjective quality evaluation is necessary to evaluate the realness, realisticness, and naturalness of deepfake media. While there has been very limited work on this topic, in 2020, Jiang et al. [ ] conducted a user study on realness of deepfake videos. They recruited 100 professional participants (most of whom are computer vision researchers), who were asked to evaluate the realness of 30 randomly selected videos from 7 deepfake video datasets (DeeperForensics-1.0, UADFV, DeepFake-TIMIT, Celeb-DF, FaceForensics++, Deep Fake Detection, and DFDC). Participants were asked to respond to the statement “The video clip looks real.” and gave scores following a five-point Likert scale (1 – clearly disagree, 2 – weakly disagree, 3 – borderline, 4 – weakly agree, 5 – clearly agree). 4.7 27 Table shows the results. Interestingly, we can see a huge difference between the realness levels of different datasets. What is probably quite surprising is that FaceForensics++, one of the most widely used deepfake datasets, has a very low MOS score and less than 9% of participants considered the 30 selected videos as real. 3 Table 3: Human-judged subjective quality (realness) of deepfake videos in 7 datasets. The MOS scores were not reported by , but calculated by us based on the raw data shown in Table 3 of [ ]. Jiang et al. 27 4.7 Discussion: Datasets Among all deepfake image and video datasets, a significant majority are about face images and videos. This is not surprising since face swapping, face attribution manipulation, and fully synthesised face images are among the hottest topics within deepfake research and real-world applications. We hope more non-face deepfake image and video datasets can be produced to support a broader range of research activities on deepfake. The subjective quality results shown in Table indicate that it is important to check realness of deep-fake media to support any performance evaluation or comparison. To ensure that the quality evaluation of datasets is fair, transparent and reliable, standard procedures need defining and a common pool of qualified human experts should be used. 3 Many authors of deepfake-related datasets attempted to classify such datasets into different generations. Chronologically speaking, we could broadly split such datasets into two generations: before 2019 and since 2019. Typically, datasets created before 2019 are relatively less advanced and smaller, while those created after 2019 tend to be larger, more diverse (i.e., covering more attributes), and of higher quality (i.e., produced by more advanced generative models). This can also be seen from the data in Table , in which the top two datasets (DeeperForensics-1 and Celeb-DF) fall within the new generation (2020), while others belong to the old generation. In addition to the two generations, a newer generation has also emerged in 2021: a number of very recent datasets started focusing on more realistic deepfakes (i.e., in the wild) or more specified areas of deepfakes (e.g., 10 focusing on multiple faces in the same video, and KoDF focusing on Korean faces). This trend shows that the deepfake research community has grown significantly in the past few years so that narrower topics have also started gaining attention and interest from some researchers. 3 FFIW K 5 Deepfake-Related Challenges, Competitions & Benchmarks This section reviews initiatives aiming to advance the state-of-the-art of detection and generation of synthetic or manipulated media (such as video, image and audio) via competitions or challenges open to the public, and ongoing benchmarks tackling specific problems. 5.1 Detection of Manipulated Media Ny fanamby fanandramana fanadihadiana (DFDC) was an initiative promoted by an AI and Media Steering Committee , including BBC, Facebook, Amazon, Microsoft and New York Times, and some universities around the world including the University of Oxford. The competition remained open from 5 September 2019 till 31 March 2020, and involved 3 stages. At first, the DFDC preview dataset was released. At a later stage, the DFDC full dataset was also made available to the 2,114 participants of the competition incorporating face and audio swap techniques for generation of deepfake content. At the final stage, the submitted models were evaluated using a test dataset (referred to as the “black box dataset”) of 10,000 videos which included deepfake videos. The best performance on the black box dataset had an accuracy of 65.18%, according to the released results [ ]. Submissions were ranked according to the overall log loss score, as defined in Eq. ( ). All top five ranked models (the winner had the lowest overall log loss) are available on GitHub. Results indicate how challenging the detection of deepfake is since the best accuracy was low and “ ”, according to Dolhansky et al. [ ]. Figure shows a screenshot of the leaderboard with the five finalists. The first top ranked model used MTCNN (Multi-tasked Cascaded Convolutional Network), the second used WS-DAN (Weakly Supervised Data Augumentation Network), and the third used the EfficientNetB7 architecture. Meta compiling the common themes observed in the winning models, they were: clever augmentations, architectures, and absence of forensics methods. Moving forward, they called for “ ”. 51 52 in-the-wild 22 53 13 many submissions were simply random 19 2 solutions that go beyond analysing images and video. Considering context, provenance, and other signals may be the way to improve deepfake detection models The Automatic Speaker Verification Spoofing And Countermeasures Challenge Workshop (ASVspoof) has been running biennially since 2015. This competition is organised by an international consortium that includes Inria and EURECOM (France), University of Eastern Finland, National Institute of Informatics (Japan), and Institute for Infocomm Research (Singapore). This year the ASVspoof challenge includes, for the first time, a sub-challenge focused on where the envisioned use case is an adversary trying to fool a human listener. The metric used for evaluating performance of submitted solutions (i.e., classifiers) is EER. Four baseline solutions (also called “countermeasures”), each using a different technique, were made available to participants with their corresponding EER metric values. The ASVspoof 2021 Speech Deepfake Database containing audio recordings with original and spoofed utterances has also been made available. The competition involves three phases : a progress phase, an evaluation phase and a post-evaluation phase; it is unclear how teams move from one phase to the next. More information about the 2021 competition is available in the published evaluation plan [ ]. The organisers of the competition noted that they opted for the EER as the performance evaluation met-ric for countermeasures submitted to the speech deepfake task for legacy reasons. They acknowledged, however, that “ ” by the ISO/IEC 19795-1:2021 standard. Despite the fact that only the 2021 ASVspoof competition contained a track explicitly related to deepfake, some data in the ASVspoof 2019 dataset (Logical Access task) used for the 2019 competition was generated using DL-based algorithms as mentioned in Section . We expect that this also holds for the ASVspoof 2021 dataset (Logical Access task). The ASVspoof 2019 competition used the EER as secondary metric; the primary performance metric used was the tandem detection cost function (t-DCF) [ ]. According to its evaluation plan [ ], t-DCF assesses the performance of the whole tandem system whereby “ ”. It is calculated according to Eq. ( ), where cm ( ) and cm( ) are, respectively, “ ”. 54 Speech DeepFake 55 56 13 EER reporting is deprecated 57 4 63 69 a CM [countermeasure] serves as a ‘gate’ to determine whether a given speech input originates from a bona fide (genuine) user, before passing it the main biometric verifier (the ASV system) 17 P s P s the miss rate and the false alarm rate of the CM system at threshold s For further information about Eq. ( , anisan'izany ny tsy tapaka 1 and 2, please refer to the ASVspoof 2019 evaluation plan [ Ny 17 C C 69 An implementation of the t-DCF metric has been made available by the ASVspoof 2019’s organisers in Python and Matlab formats. 58 59 The Face Anti-spoofing (Presentation Attack Detection) Challenge started in 2019. Its first two editions were held at the 2019 and 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2020), respectively. Its third edition was moved to be co-located with the 2021 IEEE/CVF International Conference on Computer Vision (ICCV 2021). This competition series was organised by a group of researchers from academia and industry in China, Mexico, Spain, Finland and the US. The 2021 competition was focused on 3D high-fidelity mask attacks, and followed a 2-phased Ny dingana voalohany dia ny "fase fampandrosoana"; nanomboka tamin'ny Aprily 2021 ny CASIA-SURF HiFiMask dataset was released to participants. The second phase is the “final ranking phase” (June 2021), when the competition ended. The competition adopted the following performance metrics for evaluation of the solutions submitted: attack presentation classification error rate (APCER), normal/bona fide presentation classification error rate (NPCER/BPCER), and average classification error rate (ACER), in accordance with the ISO/IEC 30107-3:2017 standard. Figure provides the leaderboard for the top three solutions. 60 61 62 63 64 3 Ny endriky ny FaceForensics is an ongoing automated benchmark for detection of face manipulation. The organisers of the benchmark made the FaceForensics++ dataset available for training. Manipulated videos (4,000 in total) were created using four techniques, i.e., two computer graphics-based approaches (Face2Face and FaceSwap) and two learning-based approaches (DeepFakes and Neural Textures). The deepfakes videos were generated using a slightly modified version of FaceSwap , and the Neural Textures videos were created using the approach proposed by Thies et al. [ ]. The benchmark test dataset is created from the collection of 1,000 images randomly selected from either the manipulation methods or the original videos [ ]. Participants have to submit results to the benchmark, rather then code like other competitions; this is illustrated in Figure Ny vokatry ny fangatahana dia aseho ao amin'ny sary , where the scores are a measure of accuracy (Eq. ( )). 65 66 61 56 4a 4b 8 The Open Media Forensics Challenge (OpenMFC, formerly DARPA MFC) is an annual image and video forensics evaluation aiming to facilitate development of multimedia manipulation detection systems. It has been organised annually starting from 2017 under the name of DARPA MFC. In 2020, the National Institute of Standards and Technology (NIST) initiated the as a new evaluation platform, based on their previous experiences with the DARPA MFC series, to make the participation more convenient for all researchers. In OpenMFC 2020, two deepfake-related tasks were included for the first time: Image GAN Manipulation Detection (IGMD) and Video GAN Manipulation Detection (VGMD). The organisers provided an image evaluation dataset for the IGMD task, containing 1,000 images from over 200 image journals , and a video evaluation dataset for the VGMD task, including over 100 test videos. Furthermore, they provided the datasets used in the previous MFC challenges as development datasets. The challenge is composed of two main phases for development and evaluation, respectively, and a pre-challenge phase for quality control testing. For evaluation of submissions, AUC-ROC is used as the primary metric. Furthermore, CDR@FAR, where CDR refers to correct detection rate or TPR (Eq. ( )) and FAR refers to false alarm rate or FPR (Eq. ( )), is also used as a metric [ ]. The DeeperForensics Challenge 2020 is a deepfake face detection challenge held at the 2020 ECCV 67 68 OpenMFC 69 70 4 5 49 71 Fanatanjahantena tsy misy olona . The challenge used the DeeperForensics1.0 dataset. 72 The organisers provided a hidden test dataset to better simulate real-world scenarios. The challenge involved two phases: the “development phase” that started in August 2020 allowing 100 successful sub-missions, and the “final test phase” that started in October 2020 allowing 2 successful submissions until the end of the month. The submissions were evaluated using the binary cross-entropy loss (BCELoss) metric, calculated according to Eq. ( ), where dia ny isan'ny lahatsary ao amin'ny fanandramana miafina, is the ground truth label of video (fake:1, real:0), and ( ) is the predicted probability that video is fake. 18 N yi i p yi i Results of the competition were discussed by Jiang et al. [ ]. The top solution used three models, i.e., EfficientNet-B0, EfficientNet-B1 and EfficientNet-B2, for classification. The second top used EfficientNet-B5 for both an image-based model and a video-based model. The third ranked solution used a 3D convolutional neural network (3DCNN). 73 26 Ny fanamby fanadihadiana amin'ny fanamafisana ny endrika 2021 is a competition hosted at the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2021). It is organised by researchers from a number of organisations in China including universities and SenseTime Research (the research arm of SenseTime , one of the major AI “unicorns” in China). The challenge aims to advance the state-of-the-art in detection of photo-realistic manipulation of images and videos. Participants are able to use a large annotated face dataset (i.e., the ForgeryNet dataset) that was obtained by applying a number of techniques for manipulation (15) and perturbation (36) to train their solutions. The phases comprise of Forgery Image Analysis, Forgery Video Analysis, Forgery Video Temporal Localization phases, and the final phase (i.e., “private test”) where participants’ models will be tested against an unseen dataset. The following metrics will be used [ ]: AUC, average precision (AP) at some “temporal Intersection over Union” (AP@tIoU) compared to a threshold ∈ [0*. ,* 0*.*95], and average recall (AR) at (AR@ ) where is the top labels returned for multi-class classifiers. 74 75 25 t 5 K K K K The 2020 CelebA-Spoof Face Anti-Spoofing Challenge was hosted at the 16 European Conference on Computer Vision (ECCV 2020). The challenge ran between August and October 2020, and aimed to advance the state-of-the-art in detecting “ ” [ ]. The organisers made the face CelebA-Spoof dataset available for the competition containing rich annotation across a range of attributes. The competition only had one phase where participants submitted their solutions to be evaluated against a test dataset; the spoof class was considered as “positive” and the live class as “negative”. Metric TPR@FPR was used and collected at three points where the TPR when FPR = 10 4 determined the final ranking. The top three finalists (see Figure ) used deep learning models ResNet, EfficientNet-B7, and a novel architecture combining Central Difference Convolutional Networks (CDCN) and Dual Attention Network (DAN). The two top ranked solutions used different strategies to boost their models’ performance: a heuristic voting scheme was used by the top-ranked solution, and a weight-after-sorting strategy was used by the second ranked solution. 76 th whether a presented face is live or spoof 76 − 5 The 2021 CSIG Challenge is the second edition of a challenge organised by the China Society of Image and Graphics . The 2021 challenge has the Fake Media Forensic Challenge as its 6 track, co-organised by CSIG’s Digital Media Forensics and Security Technical Committee and Institute of Information Engineering, Chinese Academy of Sciences . This track has two tasks, one on deepfake video detection, and the other on deepfake audio/speech detection. For the deepfake video detection task, the dataset used contains a public training set with 10,000 sound-free face videos (including 4,000 fake videos), a public test set with 20,000 face videos (the percentage of deepfake videos is unknown to participants), and a private test set that will be determined and used at the final session for selecting the winners. All videos contain faces of Eastern Asian people, and cover a wide range of parameters such as multiple resolutions and encoding quality factors, the use of blurring or sharpening filters, and added noise. Deepfake videos were created using public tools including DeepFaceLab [ ], Faceswap Ao amin'ny sehatry ny famerenana, ny famerenana, ny famerenana [ ] and ALAE (Adversarial Latent Autoencoders) [ ]. For the deepfake audio/speech detection task, the dataset used contains a public training set with 10,000 speech samples (including 6,000 fake ones), a public test set with 20,000 face videos (the percentage of deepfake videos is unknown to participants), and a private test set for the final session (the same as the deepfake video detection task). The tools used for generating the fake speech samples include TTS (text-to-speech) voice synthesis tools and VC (voice conversion) tools. The main TTS tools used include open-source tools such as DeepVoice, TensorFlowTTS and GAN-TTS [ ] and commercial software tools such as those from iFlytek and IBM. The main VC tools used include Adaptive-VC and CycleGAN-VC [ ]. For both deepfake detection tasks, the performance metric used is log loss. 77 78 79 th 80 81 53 82 6 54 83 8 84 29 2020 China Artificial Intelligence was the second edition of a Chinese AI competition open for the general public to participate, organised by the municipal government of the City of Xiamen in China. In 2020, it had two sub-competitions, Multimedia Information Recognition Technology Competition and Language and Knowledge Technology Competition Ny fifaninanana Multimedia Information Recognition Technology dia ahitana asa roa momba ny deepfakes: iray amin'ny fanadihadiana video deepfake. and one on deepfake audio/speech detection . The deepfake video detection task used 3,000 videos, and log loss was used as the sole performance metric. The deepfake audio/speech detection task used 20,000 audio samples (mostly in Chinese, and the remaining in English), and EER was used as the sole performance metric. For both tasks, the ratio between real and deepfake samples was 1:1. We did not find where to download the datasets used for the tasks nor a more detailed technical description of the datasets. For the deepfake video detection tasks, the top two winning teams (with an A prize) were from Netease (Hangzhou) Network Co., Ltd. and Beijing RealAI Technology Co., Ltd., followed by three other teams winning a B prize: Xiamen Fuyun Information Technology Co., Ltd.; Institute of Computing Technology, Chinese Academy of Sciences; and Wuhan Daqian Information Technology Co., Ltd. For the deepfake audio/speech task, there was no team winning an A prize, but one team winning a B prize: SpeakIn Technologies Co., Ltd. The final results of some teams were published, but some teams were allowed to hide their results. We did not find a detailed technical report summarising the results and explaining the work of the winning teams. 85 86 87 88 89 One of the B-prize winning team is from Beijing RealAI Technology Co., Ltd., a Chinese company active in deepfake-related R&D. 5.2 Generation of Manipulated Media The Voice Conversion Challenge is a biennial competition that has been running since 2016. The challenge and the corresponding workshop, hosted at the INTERSPEECH conference , is supported by the SynSig (Speech Synthesis Special Interest Group) of the International Speech Communication Association (ISCA) . Its aim is to promote progress in voice conversion (VC) technology that can be applied to a number of positive and negative use cases, such as spoofing voice biometric systems. The 2020 challenge focused on speaker conversion, a sub-problem of VC, and included two tasks. For the first task “intra-lingual semi-parallel voice conversion”, participants had to develop 16 VC systems (speaker-pair combinations) including male and female speakers and English sentences, using the provided Voice Conversion Challenge 2020 database v1.0 for training (refer to Section ). For the second task “cross-lingual voice conversion”, participants had to develop 24 VC systems, also including male and female speakers, but uttering sentences in three languages (Finnish, German and Mandarin), based on the provided training dataset. Figure illustrates the process of training and generation of VC systems. 90 91 92 93 4 6 Submissions were evaluated for “ ” . As such, the organisers used [ ] and recruited both native and non-native English speakers (i.e., Japanese native speakers) via crowd-sourcing for the listening tests. Naturalness (answering the question “How natural does the converted voice sound? ”) was measured using the metric MOS (covered in Section 4.6), and similarity (answering the question “how similar the converted voice sound comparing source and target speakers? ”) was measured in terms of speaker recognition as “same” or “different”, as elaborated by Wester et al. [68]. Tests also focused on the effects of language differences on the performance of VC systems submitted to the competition. The most popular CNN/RNN/GANbased VC systems submitted used WaveNet, WaveRNN, and Parallel WaveGAN. Results indicated that, in terms of similarity, the best performing VC systems were as good as natural speech but none reached human-level naturalness for task 1; scores were lower for task 2 which was more complex [70]. The organisers of the 2020 competition also used objective evaluation [12]. The metrics used for evaluation of speaker similarity were: equal error rate (EER), false acceptance rate of target (P tar fa ), miss rate of source (P src miss), and cosine similarity of speaker embedding vectors (cos-sim) according to Eq. (19) where A is the speaker embedding vectors for the converter audio and B is the speaker embedding vectors for the original audio. The performance of the VC systems as a spoof countermeasure was also evaluated using EER, while to evaluate the quality of the subjective MOS obtained via listening tests, a DL-based model to predict MOS, called MOSNet [43], was used. Lastly, to evaluate intelligibility of the converted transcribed speech, in comparison with the original transcribed speech, the word error rate (WER) [4] was used. WER is calculated according to Eq. (20) where I refers to insertions, D refers to deletions, S refers to substitutions, and N refers to the total number of words in the original transcript. perceived naturalness and similarity through listening tests 94 Ny fanombanana 70 The Deepfake Africa Challenge (2021) is a new initiative of the AI Africa Expo, in partnership with a film and media production company (Wesgro) and the African Data Science competition platform Zindi. Its aim is “ ”. Eligible participants were required to be citizens and residents of the African continent. Submissions, accepted up to end of July 2021, can be either video or audio. Evaluation of submissions is defined in terms of artistic creativity, relevance of challenge topic, and innovation in the process of generation as long as participants use tools and packages publicly available. The top three finalists will receive a prize, present their work at the Expo, and will have to grant copyrights to Zindi. Unlike the other competitions reviewed in this section, which were focused on advancing the state-of-the-art in detection of synthetic or manipulated media, this competition focused on the generation of deepfake which seems more humanities-centred. This is a trend observed in arts [ ] and culture [ ].5.3 Generation and Detection of Manipulated Media 95 to create convincing deepfakes to highlight the power of this synthetic media, illustrating its creative potential for exploitation for both positive and negative outcomes and focusing debate about its ethical use / misuse in an African context 31 57 The DeepFake Game Competition (DFGC) Ao amin'ny 2021 International Joint Conference on Biometrics (IJCB 2021). Ny mpikarakara azy dia indrindra avy amin'ny Institute of Automation Chinese Academy of Sciences (CASIA). Ny hevitra momba ny fifaninanana dia ny hampitombo ny lalao mifanohitra eo amin'ny mpiara-miasa miantso ny fandrosoana eo amin'ny famoronana sy ny fametrahana sy ny famantarana. Mba hahatratrarana izany, ny protocol 6 dingana dia novolavolaina miaraka amin'ny dingana telo famoronana (C-fase) sy ny dingana famantarana (D-fase), matetika herinandro iray hafa; ny fangatahana dia tapitra tamin'ny Aprily 2021. Ny dingana C sy ny dingana D dia mifandray amin'ny rakitra Celeb-DF (v2) [ ], containing 6,229 videos (590 real/original videos and 5,639 fake/manipulated videos), for training purposes. As such, submissions to a C-phase would consist of datasets extracted from Celeb-DF (v2) which included novel face-swap approaches to obtain evaluation results. Submissions to a D-phase would consist of detection models/codes to obtain evaluation results. The models submitted for a D-phase were evaluated against the datasets submitted for the previous C-phase [ ]. The metrics used for evaluation were: a detection score, used for evaluation of a D-phase, and a creation score, used for evaluation of a C-phase. The top three finalists for the detection phase employed CNN-based classifiers EfficientNet-B3, Efficientnet-B0 and EfficientNetV2. 96 40 52 97 The Detection Score ( ) metric captures the models’ ability to correctly classify fake images submitted to the previous C-phase against a set of real images in the CelebDF test dataset. It is calculated using Eq. ( ), where dia ny isan'ny fangatahana manan-kery amin'ny fanandramana synthesis natsangana tao amin'ny dingana C farany. Ny 21 NC The Creation Score ( ) metric used to evaluate creation models submitted to this challenge is calculated by Eq. ( Aiza ny is the number of valid submissions of detection methods in the last D-phase, the noise score ( noise) penalises noisy images, the other three parts of the equation relate to the following : “ ”. CS 22 ND S 98 ID level similarity to the donor ID, image level similarity to the target frame, and the deception ability against detection models. ID level similarity is scored by a face recognition model using dot product of two ID features (fake face ID and donor ID). The image level similarity is scored by SSIM [Structural Similarity Index] to make sure the face-swapped image is similar to the corresponding target image in content and quality Peng et al. [ ] observed a commonality between the three winning teams for the creation task, i.e., the use of the FaceShifter [ ] framework for face swapping. They highlighted two overall reflections about the competition: (1) the limited diversity of the deepfake datasets submitted and the use of repetitive methods to generate them, and (2) the limited size of the Celeb-DF (v2) dataset itself flagging the need for a larger dataset for next year’s competition. The organisers of the competition also applied the top two detection models to unseen datasets (DFDC and FaceForensics++) and noticed that they do not generalise well. 52 37 6 A Meta-Review of Deepfake-Related Surveys This section presents a meta-review of 12 selected deepfake-related survey papers, including eight published in English [ , , , – , , ] and four published in Chinese [ , , , ]. It covers the following aspects in a systematic manner: definitions and scope, performance metrics, datasets, challenges/competitions/benchmarks, performance comparison, key challenges and recommendations. 16 45 46 64 66 71 73 7 38 41 59 The meta-review aims at drawing some high-level insights for monitoring future development of deepfake-related technologies and their applications. 6.1 Definitions and Scope As we discussed in Section , eo amin'ny mpikaroka, matihanina sy ny mpikatroka lalàna dia tsy misy famaritana ankapobeny ny "deepfake" ho fehezan-teny. Izany dia hita ihany koa ao amin'ny fomba ny mpanoratra ny 12 tatitra fanadihadiana nahita ity lafiny ity. Ny ankamaroan'ny mpanoratra niteny momba ny tantaran'ny deepfakes ary nilaza fa ny teny dia maneho ny fiaraha-miombon'ny "deep learning" sy "fake", fa ny sasany dia nampiasa famaritana lehibe kokoa, ohatra, Lyu [ ] defined deepfake as “ ”. Some authors also referred to deepfake-related legislations, but none of them pointed out that the definitions in some such legislations are completely different from the more technical definitions involving the use of deep learning. No authors discussed the blurred boundary between deepfakes and non-deepfakes, although some surveys actually cover both, e.g., Tao et al. [ ] focused on speech forgery and did not explicitly highlight “deepfake”. 1.1 45 high quality fake videos and audios generated by AI algorithms 59 In terms of the scope, while some authors (correctly) considered all types of media that can be produced by deepfake-related techniques [ , , , ], some considered only a narrow scope, e.g., authors of [ , , Ny ] considered only videos, and only authors of [ , ] have considered images and videos. Another phenomenon we observed is that many authors focused more on face images and videos, and authors of three surveys [ , Ny ] na dia voafetra ny famaritana ny "deepfake" ho amin'ny fifehezana mahazatra toy izany: 38 41 45 65 7 64 71 73 16 66 16 64 71 Deshmukh and Wankhade [ ] defined it as “ ”; 16 a technology which creates fake images or videos of targeted humans by swapping their faces [by] another character saying or doing things that are not absolutely done by them and humans start believing in such fake as it is not always recognisable with the everyday human eye Younus and Hasan [ ] considered deepfake as a technique allowing “ ”; and 71 any computer user to exchange the face of one person with another digitally in any video Tolosana et al. [ ] defined it as “ ”. 64 a deep learning based technique able to create fake videos by swapping the face of a person by the face of another person Such unnecessarily narrow definitions and scopes can lead to confusion and do not help exchanges between researchers and practitioners working on different types of deepfakes. Miantso ireo mpikaroka bebe kokoa isika mba handray ny famaritana lehibe kokoa ny "deepfake" mba hahatonga ny haino aman-jery tena marina / voajanahary amin'ny karazana na inona na inona niteraka amin'ny alalan'ny fomba mahomby automatique (matetika mifototra amin'ny AI) dia heverina ho deepfake. eto, manome ohatra roa toy izany famaritana lehibe kokoa: ny image2image (na pixel2pixel) teknika [ ] that allows the production of deepfake images and videos of any objects (e.g., the “horse2zebra” deepfake image shown in Figure ), and the the so-called “deepfake geography [ ]”, where AI-based techniques are used to generate realistic-looking satellite images. 80 7 77 Another important fact missed or not sufficiently discussed by authors of all the 12 surveys is that deepfake techniques can be used for positive applications, e.g., creative arts, entertainment and protecting online users’ privacy. We call for more researchers and practitioners to follow the proposal in the 2020 Tencent AI White Paper [ ] to start using the more neutral-sounding term “deep synthesis”. Accordingly,we can use different words for different types of data generated using “deep synthesis” techniques, e.g., “deep art”, “deep animation”, “deep music”, and “deepfake”. While authors of the 12 survey papers did not recognise the positive applications of “deepfake” technologies, some other researchers did, e.g., organisers of the Voice Conversion Challenge 2020 who said the VC technology (for speech deepfake) “ ”. 60 99 is useful in many applications, such as customizing audio book and avatar voices, dubbing, movie industry, teleconferencing, singing voice modification, voice restoration after surgery, and cloning of voices of historical persons 6.2 Performance Metrics Surprisingly, none of the 12 surveys have covered performance metrics explicitly. Some directly used performance metrics to explain and compare performance of covered deepfake generation and detection methods. The most used performance metrics include accuracy, ERR, and AUC. This may be explained by the page constraints of such survey papers, which did not allow the authors to extend their coverage significantly to cover performance metrics systematically. The subjective quality of deepfakes is an area least covered by the surveys, which seems related to an unbalanced coverage on deepfake generation and deepfake detection in terms of performance evaluation and comparison (the former much less than the latter). 6.3 Datasets Maro amin'ireo tatitra 12 momba ny fanadihadiana dia mamaritra ny isan'ny angon-drakitra mifandraika amin'ny deepfake, saingy tsy misy amin'izy ireo manana fitehirizana tanteraka toy ny antsika hita ao amin'ny fizarana . For instance, none of the surveys have covered the Voice Conversion Challenge 2016/2018/2020 datasets and the ASVspoof 2019/2021 datasets are covered briefly only in two surveys [ , ]. In addition, more recent deepfake datasets especially those released in 2021 are also not covered by any of the surveys. We believe that our Section is the most comprehensive review of deepfake-related datasets so far. 4 38 59 4 Some survey papers include datasets that are likely deepfakes, e.g., Verdoliva [ Niharan'ny angon-drakitra ankapobeny maromaro amin'ny endriky ny sary izay tsy niteraka tamin'ny alalan'ny fampianarana maivana na ny fomba mifototra amin'ny AI, ary ny fanadihadiana sasany (ohatra, [ ]) mentioned ASVspoof 2015 datasets but we did not see the use of deep learning for generating data used in the dataset. 66 38 6.4 Challenges, Competitions and Benchmarks Many surveys cover deepfake-related challenges, competitions and benchmarks. The coverage is, however, mostly limited, and some challenges (e.g., the Voice Conversion Challenge 2016/2018/2020 and the two Chinese challenges we covered in Section ) are not covered by any of the surveys. The level of detail of challenges, competitions and benchmarks is also normally limited, compared with what we chose to include in Section . Similar to the datasets we covered in Section , we believe that our coverage of deepfake-related challenges, competitions and benchmarks in Section is also the most comprehensive so far. 5 5 4 5 6.5 Performance Comparison Most surveys have a good coverage of related methods for deepfake generation and detection, but only some explicitly covered performance comparison between different methods [ , , ]. 38 46 64 Among all the survey papers, Li et al. [ ] conducted the most comprehensive study on performance of different deepfake detection methods. In addition to showing the performance metrics of a number of deepfake detection methods in Table 3 of [ ], they also looked at general characteristics and issues of different types of deepfake detection methods, as shown in Table . Furthermore, they also looked at research on robustness of deepfake detection methods against adversarial samples, referring to some work that showed a lack of such robustness. 38 38 4 Due to quality issues of many deepfake-related datasets (discussed in Section ), we need to treat any performance metrics and comparison of different detection methods with caution. Without testing all methods on a sufficiently large, diverse and high-quality deepfake dataset, the performance comparison results can be misleading. This highlights the importance of having more challenges, competitions and benchmarks to encourage performance comparison on standard datasets and using consistent performance metrics. 4.6 6.6 Challenges and Recommendations The authors of some surveys identified some key challenges and future research directions for the deepfake community. Tsy mahagaga, ny fomba hanatsarana fomba mahery vaika, azo ampiharina, azo ampiharina ary azo aseho amin'ny fomba fanadihadiana dia iray amin'ireo fanamby fototra notontosaina indrindra ary ihany koa ny fikarohana lehibe amin'ny hoavy. [ , , , , Ny , , , ]. Considering the arms race between deepfake generation and detection, this research direction will likely remain the hottest topic in deepfake research. 7 16 38 41 45 59 65 66 71 A couple of surveys [ , ] niresaka momba ny fusion toy ny fikarohana manan-danja amin'ny hoavin'ny fikarohana, izay ny "fusion" dia mifandray amin'ny fampidirana fomba samihafa (ohatra, mampifandray detectors maro amin'ny karazana samihafa) sy ny angon-drakitra loharanom-baovao (ohatra, miara-miasa amin'ny fanadihadiana audio-visual) mba hahatratrarana ny vokatra tsara kokoa ho an'ny deepfake fikarohana. ] suggested that, for detection of deepfake videos, we need to consider video-level detection more, which can be considered fusion of detection results of all video frames. 38 66 45 The authors of three surveys, Lyu [ ] , Deshmukh and Wankhade [ [Na: Fa inona no fanantenan'ny mpihatsaravelatsihy, na dia mahazo harena be aza. ], argued that better (higher-quality, more up-to-date, and more standard) deepfake datasets are needed to develop more effective deepfake detection methods. Lyu [ Nilaza ihany koa izy fa mila mihevitra ny effects in training data and improve the evaluation of datasets. We agree with them on these points. 45 16 71 45 Fandefasana media sosialy amin'ny aterineto Tao et al. [ ] suggested that low-cost deepfake generation/detection should be considered as a future research direction. This is a valid recommendation since lightweight methods will allow less powerful computing devices (e.g., IoT devices) to benefit from such technologies. 59 Two Chinese surveys [ , ] also mentioned the need to have new deepfake-related legislations combating malicious use of deepfakes and the need to train end users such as journalists. This is likely an area where interdisciplinary research can grow. 38 41 There are also other ad-hoc recommendations given by the authors of some surveys. For example, Lyu [ ] argued that deepfake detection should be considered a (more complicated) multi-class, multi-label and local detection problem. Tolosana et al. [ Niresaka momba ny fikarohana manokana ho an'ny fomba samihafa amin'ny famokarana lalina (fifandraisana synthesis, ny identity swap, ny fanitsiana ny toetra, ary ny fanehoan-kevitra swap). [Na: Fa inona no fanantenan'ny mpihatsaravelatsihy? ] recommended more active defence mechanisms such as using digital watermarking and blockchain technologies to build trustworthy media frameworks against deepfakes. 45 64 41 38 7 Conclusion The rapid growth in the capability to manipulate media or create synthetic media which look realistic and natural paved the way for deepfakes. At first, this paper adopted a critical approach to look at different definitions of the term “deepfake”. In that regard, we point out the different contradicting definitions and call for the wider community to consider how to define a new term that has a more consistent scope and meaning. For instance, replacing “deepfake” by “deep synthesis” can be more inclusive by embracing positive applications of deepfake techniques, e.g., in entertainment and for simulation purposes. Ity lahatsoratra ity dia nanome fanadihadiana feno momba ny lafiny maro amin'ny tontolo iainana sy ny tsiambaratelon'ny fikarohana sy ny loharano an-tserasera hafa izay navoaka amin'ny fiteny roa: teny Anglisy sy Shinoa. Ity lahatsoratra ity dia ahitana ny fampisehoana sy ny fitsipika ampiasaina matetika, ny angon-drakitra mifandraika, ny fanamby, ny fifaninanana ary ny tsiambaratelo. Ny References [1] Darius Afchar, Vincent Nozick, Junichi Yamagishi, and Isao Echizen. 2018. MesoNet: A Compact Facial Video Forgery Detection Network. In . IEEE, 1–7. Proceedings of the 2018 IEEE International Workshop on Information Forensics and Security https://doi.org/10.1109/WIFS.2018.8630 761 [2] Henry Ajder, Giorgio Patrini, Francesco Cavalli, and Laurence Cullen. 2019. The State of Deepfakes: Landscape, Threats, and Impact. Deeptrace. , 27 pages. https://sensity.ai/reports/ [3] Zahid Akhtar and Tiago H. Falk. 2017. Audio-Visual Multimedia Quality Assessment: A Comprehensive Survey. 5 (2017), 21090–21117. IEEE Access https://doi.org/10.1109/ACCESS.2017.2750918 [4] Ahmed Ali sy Steve Renals. 2018. Word Error Rate Estimation for Speech Recognition: e-WER. . Association for Computational Linguistics, 20–24. Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics https://doi.org/10.18653/v1/P18-2004 [5] Sercan O¨ . Arık, Jitong Chen, Kainan Peng, Wei Ping, and Yanqi Zhou. 2018. Neural Voice Cloning with a Few Samples. In . Curran Associates Inc., 10040–10050. Proceedings of the 32nd International Conference on Neural Information Processing Systems https://papers.nips.cc/paper/201 8/hash/4559912e7a94a9c32b09d894f2bc3c82-Abstract.html [6] Aayush Bansal, Shugao Ma, Deva Ramanan, and Yaser Sheikh. 2018. Recycle-GAN: Unsupervised Video Retargeting. In Torohevitra 17 amin'ny pejy. Proceedings of the 2018 European Conference on Computer Vision https://doi.org/10.1007/978-3-030-01228-1 8 [7] Yu-xuan Bao, Tian-liang Lu, and Yan-hui Du. 2020. Overview of Deepfake Video Detection Technology. 47, 9 (2020), 283–292. Computer Science https://doi.org/10.11896/jsjkx.200400130 [8] Mikol-aj Bin´kowski, Jeff Donahue, Sander Dieleman, Aidan Clark, Erich Elsen, Norman Casagrande, Luis C. Cobo, and Karen Simonyan. 2019. High Fidelity Speech Synthesis with Adversarial Networks. https://doi.org/10.48550/ARXIV.1909.11646 [9] ![](file:///C:/Users/user/AppData/Local/Temp/msohtmlclip1/01/clip_image050.gif)Madeline Brady. 2020. Deepfakes: A New Desinformation Threat? Report by the Democracy Reporting International. , 9 pages. https://democracy-reporting.org/dri publications/de epfakes-a-new-disinformation-threat/ [10] Umur Aybars Ciftci, Ilke Demir, ary Lijun Yin. 2020. FakeCatcher: Fahafantarana ny lahatsary syntetika mampiasa famantarana ara-biolojika. (2020), 17 pages. IEEE Transactions on Pattern Analysis and Machine Intelligence https://doi.org/10.1109/TPAMI.2020.3009287 [11] Hao Dang, Feng Liu, Joel Stehouwer, Xiaoming Liu, and Anil K. Jain. 2020. On the Detection of Digital Face Manipulation. In . IEEE, 10 pages. Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition https://doi.org/10.1109/CVPR42600.2020.00582 [12] Rohan Kumar Das, Tomi Kinnunen, Wen-Chin Huang, Zhen-Hua Ling, Junichi Yamagishi, Zhao Yi, Xiaohai Tian, and Tomoki Toda. 2020. Predictions of Subjective Ratings and Spoofing Assessments of Voice Conversion Challenge 2020 Submissions. In . International Speech Communication Association, 99–120. Proceedings of the Joint Workshop for the Blizzard Challenge and Voice Conversion Challenge 2020 https://doi.org/10.21437/VCC BC.2020-15 [13] H´ector Delgado, Nicholas Evans, Tomi Kinnunen, Kong Aik Lee, Xuechen Liu, Andreas Nautsch, Jose Patino, Md Sahidullah, Massimiliano Todisco, Xin Wang, and Junichi Yamagishi. 2021. ASVspoof 2021: Automatic Speaker Verification Spoofing and Countermeasures Challenge Evaluation Plan. https://www.asvspoof.org/asvspoof2021/asvspoof2021 evaluation plan.pdf [14] H´ector Delgado, Nicholas Evans, Tomi Kinnunen, Kong Aik Lee, Xuechen Liu, Andreas Nautsch, Jose Patino, Md Sahidullah, Massimiliano Todisco, Xin Wang, and Junichi Yamagishi. 2021. ASVspoof 2021 Challenge - Logical Access Database. https://doi.org/10.5281/zenodo.4837263 [15] H´ector Delgado, Nicholas Evans, Tomi Kinnunen, Kong Aik Lee, Xuechen Liu, Andreas Nautsch, Jose Patino, Md Sahidullah, Massimiliano Todisco, Xin Wang, and Junichi Yamagishi. 2021. ASVspoof 2021 Challenge - Speech Deepfake Database. https://doi.org/10.5281/zenodo.4835108 [16] ![](file:///C:/Users/user/AppData/Local/Temp/msohtmlclip1/01/clip_image051.gif)Anushree Deshmukh sy Sunil B. Wankhade. 2021. Deepfake Detection Approaches Using Deep Learning: A Systematic Review. In . Springer, 293–302. Intelligent Computing and Networking: Proceedings of IC-ICN 2020 (Ny fampianarana amin'ny tambajotra sy ny rafitra, Vol. 146) https://doi.org/ 10.1007/978-981-15-7421-4 27 [17] Xinyi Ding, Zohreh Raziei, Eric C. Larson, Eli V. Olinick, Paul Krueger, and Michael Hahsler. 2020. Swapped Face Detection using Deep Learning and Subjective Assessment. 2020, 1 (2020), 1–12. EURASIP Journal on Information Security https://doi.org/10.1186/s13635-020-00109-8 [18] Brian Dolhansky, Joanna Bitton, Ben Pflaum, Jikuo Lu, Russ Howes, Menglin Wang, and Cristian Canton Ferrer. 2020. The DeepFake Detection Challenge (DFDC) Dataset. https://doi.org/10.48550/ARXIV.2006.07397 [19] Brian Dolhansky, Joanna Bitton, Ben Pflaum, Jikuo Lu, Russ Howes, Menglin Wang, ary Cristian Canton Ferrer. 2020. The DeepFake Detection Challenge (DFDC) Dataset. arXiv:2006.07397. https://arxiv.org/abs/2006.07397 [20] Nick Dufour and Andrew Gully. 2019. Contributing Data to Deepfake Detection Research. Google AI Blog. https://ai.googleblog.com/2019/09/contributing-data-to-deepfake-detectio Ny HTML [21] Ricard Durall, Margret Keuper, Franz-Josef Pfreundt, and Janis Keuper. 2019. Unmasking Deep-Fakes with Simple Features. https://doi.org/10.48550/ARXIV.1911.00686 [22] Cristian Canton Ferrer, Brian Dolhansky, Ben Pflaum, Joanna Bitton, Jacqueline Pan, and Jikuo Lu. 2020. Deepfake Detection Challenge Results: An Open Initiative to Advance AI. Meta AI Blog. https://ai.facebook.com/blog/deepfake-detection-challenge-results-an-open-initia tive-to-advance-ai/ [23] Gereon Fox, Wentao Liu, Hyeongwoo Kim, Hans-Peter Seidel, Mohamed Elgharib, and Christian Theobalt. 2021. Videoforensicshq: Detecting High-Quality Manipulated Face Videos. In . IEEE, 1–6. Proceedings of the 2021 IEEE International Conference on Multimedia and Expo https://doi.or g/10.1109/ICME51207.2021.9428101 [24] Haiying Guan, Andrew Delgado, Yooyoung Lee, Amy N. Yates, Daniel Zhou, Timothee Kheyrkhah, and Jon Fiscus. 2021. User Guide for NIST Media Forensic Challenge (MFC) Datasets. https://doi.org/10.6028/NIST.IR.8377 [25] Yinan He, Bei Gan, Siyu Chen, Yichun Zhou, Guojun Yin, Luchuan Song, Lu Sheng, Jing Shao, ary Ziwei Liu. 2021. ForgeryNet: Benchmark isan-karazany ho an'ny fanadihadiana ankapobeny. . IEEE, 4360–4369. Ny fifanarahana IEEE / CVF 2021 momba ny fahitana ny solosaina sy ny famantarana ny endrika https://doi.org/10.1109/CVPR46437.2021.00434 [26] Liming Jiang, Zhengkui Guo, Wayne Wu, Zhaoyang Liu, Ziwei Liu, Chen Change Loy, Shuo Yang, Yuanjun Xiong, Wei Xia, Baoying Chen, Peiyu Zhuang, Sili Li, Shen Chen, Taiping Yao, Shouhong Ding, Jilin Li, Feiyue Huang, Liujuan Cao, Rongrong Ji, Changlei Lu, and Ganchao Tan. 2021. DeeperForensics Challenge 2020 on Real-World Face Forgery Detection: Methods and Results. arXiv:2102.09471. https://arxiv.org/pdf/2102.09471.pdf [27] Liming Jiang, Ren Li, Wayne Wu, Chen Qian, and Chen Change Loy. 2020. DeeperForensics-1.0: A Large-Scale Dataset for Real-World Face Forgery Detection. In . IEEE, 2886–2895. Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition https://doi.org/ 10.1109/CVPR42600.2020.00296 [28] Nal Kalchbrenner, Erich Elsen, Karen Simonyan, Seb Noury, Norman Casagrande, Edward Lockhart, Florian Stimberg, Aaron van den Oord, Sander Dieleman, and Koray Kavukcuoglu. 2018. Efficient Neural Audio Synthesis. https://doi.org/10.48550/ARXIV.1802.08435 [29] Takuhiro Kaneko and Hirokazu Kameoka. 2017. Parallel-Data-Free Voice Conversion Using Cycle-Consistent Adversarial Networks. https://doi.org/10.48550/ARXIV.1711.11293 [30] Tero Karras, Samuli Laine, and Timo Aila. 2019. A Style-based Generator Architecture for Generative Adversarial Networks. In . IEEE, 4401–4410. Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition https://doi.org/10.1109/CVPR.2019.00453 [31] Ross Kelly. 2021. Digital Art Exhibition to Showcase Creative Potential of AI. DIGIT News. https://www.digit.fyi/digital-art-exhibition-to-showcase-creative-potential-of-ai/ [32] Ali Khodabakhsh, Raghavendra Ramachandra, Kiran Raja, Pankaj Wasnik, and Christoph Busch. 2018. Fake Face Detection Methods: Can They Be Generalized?. In . IEEE, 1–6. Proceedings of the 2018 International Conference of the Biometrics Special Interest Group https://doi.org/10.2 3919 / BIOSIG.2018.8553251 Ny fampiasana [33] Hyeongwoo Kim, Mohamed Elgharib, Hans-Peter Zoll ̈ofer, Michael Seidel, Thabo Beeler, Christian Richardt, ary Christian Theobalt. 38, 6, Andininy 178 (2019), 13 pejy. ACM Transactions on Graphics https://doi.org/10.1145/3355089.3356500 [34] Hyeongwoo Kim, Pablo Garrido, Ayush Tewari, Weipeng Xu, Justus Thies, Matthias Niessner, Patrick P´erez, Christian Richardt, Michael Zollh¨ofer, and Christian Theobalt. 2018. Deep Video Portraits. 37, 4, Article 163 (2018), 14 pages. ACM Transactions amin'ny Graphics https://doi.org/ 10.1145/3197517.3201283 [35] Pavel Korshunov sy S'ebastien Marcel. 2019. Fanombanana ny fahavoazana sy ny fanadihadiana ny lahatsary maloto. Ny isa dia 1 - 6. Fifanarahana iraisam-pirenena momba ny Biometrika 2019 https://doi.org/10.1109/ICB45273.2019.8987375 [36] Patrick Kwon, Jaeseong You, Gyuhyeon Nam, Sungwoo Park, and Gyeongsu Chae. 2021. KoDF: A Large-scale Korean DeepFake Detection Dataset. In . IEEE, 10724–10733. Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision https://doi.org/10.1109/ICCV Ny laharam-pahamehana amin'izao fotoana izao [37] Lingzhi Li, Jianmin Bao, Hao Yang, Dong Chen, and Fang Wen. 2020. FaceShifter: Towards High Fidelity And Occlusion Aware Face Swapping. arXiv:1912.13457. https://arxiv.org/abs/1912.13457 [38] Xurong Li, Shouling Ji, Chunming Wu, Zhenguang Liu, Shuiguang Deng, Peng Cheng, Min Yang, and Xiangwei Kong. 2021. Survey on Deepfakes and Detection Techniques. 32, 2 (2021), 496–518. Journal of Software http://www.jos.org.cn/1000-9825/6140.htm [39] Yuezun Li, Ming-Ching Chang, ary Siwei Lyu. 2018. Ao amin'ny Ictu Oculi: Ny fanambaran'ny AI dia namorona lahatsary ratsy amin'ny alàlan'ny fanadihadiana ny mijery ny masony. ao Izany dia ny 1 - 7. Proceedings of the 2018 IEEE International Workshop on Information Forensics and Security https://doi.org/10.1109/WIFS.2018.8630787 [40] Yuezun Li, Xin Yang, Pu Sun, Honggang Qi, and Siwei Lyu. 2020. Celeb-DF: A Large-Scale Challenging Dataset for DeepFake Forensics. In . IEEE, 3204–3213. Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition https://doi.org/10.1109/CVPR42600. 2020.00327 [41] Ruigang Liang, Peizhuo Lv, Yue Zhao, Peng Chen, Hao Xing, Yingjun Zhang, Jizhong Han, Ran He, Xianfeng Zhao, Ming Li, ary Kai Chen. 5, 2 (2020), 1–17. Journal of Cyber Security http://jcs.iie.ac.cn/xxaqxb/ch/re ader/view abstract.aspx?file no=20200202&flag=1 [42] Steven R. Livingstone and Frank A. Russo. 2018. The Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS): A Dynamic, Multimodal Set of Facial and Vocal Expressions in North American English. 13, 5 (2018), 35 pages. PloS one [43] Chen-Chou Lo, Szu-Wei Fu, Wen-Chin Huang, Xin Wang, Junichi Yamagishi, Yu Tsao, and Hsin-Min Wang. 2021. MOSNet: Deep Learning based Objective Assessment for Voice Conversion. arXiv:1904.08352. https://arxiv.org/pdf/1904.08352.pdf [44] Jaime Lorenzo-Trueba, Junichi Yamagishi, Tomoki Toda, Daisuke Saito, Fernando Villavicencio, Tomi Kinnunen, ary Zhenhua Ling. 2018. The Voice Conversion Challenge 2018: Fanentanana ny fampandrosoana ny fomba mitovy sy tsy mitovy. . International Speech Communication Association, 195–202. Proceedings of the Odyssey 2018 The Speaker and Language Recognition Workshop https://doi.org/10.21437/Odyssey.2018-28 [45] Siwei Lyu. 2020. Deepfake Detection: Current Challenges and Next Steps. In . IEEE, 6 pages. Ny fandaharam-potoana amin'ny 2020 IEEE International Conference on Multimedia Expo Workshops https://doi.org/10.1109/ICMEW46912.2020.9105991 [46] Yisroel Mirsky and Wenke Lee. 2021. The Creation and Detection of Deepfakes: A Survey. 54, 1, Article 7 (2021), 41 pages. ACM Computing Survey https://doi.org/10.1145/3425780 [47] Gautham J. Mysore. 2015. Can we Automatically Transform Speech Recorded on Common Consumer Devices in Real-World Environments into Professional Production Quality Speech?—A Dataset, Insights, and Challenges. 22, 8 (2015), 1006–1010. IEEE Signal Processing Letters https://doi.org/10.1109/LSP.2014.2379648 [48] Jo˜ao C. Neves, Ruben Tolosana, Ruben Vera-Rodriguez, Vasco Lopes, Hugo Proen¸ca, and Julian Fierrez. 2020. GANprintR: Improved Fakes and Evaluation of the State of the Art in Face Manipulation Detection. 14, 5 (2020), 1038 hatramin'ny 1048 IEEE Journal of Selected Topics amin'ny famoahana famantarana https://doi.org/10.1109/JSTSP.2020.3007250 [49] NIST Media Forensics Challenge Team. 2021. Open Media Forensics Challenge 2020-2021 Evaluation Plan. https://mig.nist.gov/MFC/Web/EvalPlan2020/OpenMFC2020EvaluationPlan.pdf [50] Aaron van den Oord, Sander Dieleman, Heiga Zen, Karen Simonyan, Oriol Vinyals, Alex Graves, Nal Kalchbrenner, Andrew Senior, ary Koray Kavukcuoglu. https://doi.org/10.48550/ARXIV.1609.03499 [51] Debajyoti Pal and Tuul Triyason. 2018. A Survey of Standardized Approaches towards the Quality of Experience Evaluation for Video Services: An ITU Perspective. 2018, Article 1391724 (2018), 25 pages. International Journal of Digital Multimedia Broadcasting https://doi.org/10.1155/20 18 taona 17179 [52] Bo Peng, Hongxing Fan, Wei Wang, Jing Dong, Yuezun Li, Siwei Lyu, Qi Li, Zhenan Sun, Han Chen, Baoying Chen, Yanjie Hu, Shenghai Luo, Junrui Huang, Yutong Yao, Boyuan Liu, Hefei Ling, Guosheng Zhang, Zhiliang Xu, Changtao Miao, Changlei Lu, Shan He, Xiaoyan Wu, and Wanyi Zhuang. 2021. DFGC 2021: A DeepFake Game Competition. arXiv:2106.01217. https: //arxiv.org/abs/2106.01217 [53] Ivan Perov, Daiheng Gao, Nikolay Chervoniy, Kunlin Liu, Sugasa Marangonda, Chris Um ́e, Mr. Dpfks, Carl Shift Facenheim, Luis RP, Jian Jiang, Sheng Zhang, Pingyu Wu, Bo Zhou, ary Weiming Zhang. 2020. DeepFaceLab: Integrated, Flexible and Extensible Face-swapping Framework. https://doi.org/10.48550/ARXIV.2005.05535 [54] Stanislav Pidhorskyi, Donald A. Adjeroh, ary Gianfranco Doretto. 2020. Adversarial Latent Au-to-coders. Ao amin'ny . IEEE, 10 pages. Fifanarahana IEEE/CVF 2020 momba ny fahitana sy ny famantarana endrika amin'ny solosaina https://doi.org/10.1109/CVPR42600.2020.01411 [55] Andreas R ̈ossler, Davide Cozzolino, Luisa Verdoliva, Christian Riess, Justus Thies, ary Matthias Nießner. 2018. FaceForensics: Dataset lahatsary lehibe ho an'ny fanadihadiana ny fandikan-teny amin'ny endrika olombelona. https://doi.org/10.48550/ARXIV.1803.09179 [56] Andreas R¨ossler, Davide Cozzolino, Luisa Verdoliva, Christian Riess, Justus Thies, and Matthias Nießner. 2019. FaceForensics++: Learning to Detect Manipulated Facial Images. In . IEEE, 1–11. Fifanarahana iraisam-pirenena momba ny fahitana amin'ny solosaina 2019 https://doi.org/10.110 9/ICCV.2019.00009 [57] Anindya Sen. 2021. Art and Artificial Intelligence: How ‘Deepfakes’ Can Help Create Authentic Museum Experiences! https://medium.com/art-world-zen/art-and-artificia l-intelligence-how-deepfakes-can-help-create-authentic-museum-experiences-65d8aa 7da29c [58] Kou Tanaka, Hirokazu Kameoka, Takuhiro Kaneko, ary Nobukatsu Hojo. 2019. WaveCycleGAN2: Time-domain Neural Post-filter for Speech Waveform Generation. https://doi.org/10.48550/A RXIV 1904.02892 amin'ny alalan'ny [59] Jianhua Tao, Ruibo Fu, Jiangyan Yi, Chenglong Wang, ary Tao Wang. 2020. fampandrosoana sy fanamby ny fitenenana sy ny famantarana ny teny. 5, 2 (2020), 28–38. Gazety momba ny fiarovana cyber http://jcs.iie.ac.cn/xxaqxb/ch/reader/view abstract.aspx?file no=20200204&flag=1 [60] Tencent. 2020 Intellectual Intelligence White Paper momba ny fikarohana ara-teknika. https://tech.sina.com.cn/roll/2020-07-14/doc-iivhvpwx5201226.shtml [61] Justus Thies, Michael Zollh ̈ofe, ary Matthias Niessner. 2019. Ny fametrahana ny neurons ho voafetra: Synthesis ny sary amin'ny fampiasana ny Textures Neural. 38 Ny andininy 66 (2019), 12 pejy. ACM Transactions amin'ny Graphics https://doi.org/10.1145/3306346.3323035 [62] Tomoki Toda, Ling-Hui Chen, Daisuke Saito, Fernando Villavicencio, Mirjam Wester, Zhizheng Wu, ary Junichi Yamagishi. 2016. Ny fanamby fanovàna feo 2016. Ny fikambanana iraisam-pirenena momba ny fifandraisana amin'ny teny, 1632-1636. Ny fandaharam-potoana amin'ny Interspeech 2016 https://doi.org/10.21437/Interspeech.2016-1066 [63] Massimiliano Todisco, Xin Wang, Ville Vestman, Md Sahidullah, Hector Delgado, Andreas Nautsch, Junichi Yamagishi, Nicholas Evans, Tomi Kinnunen, and Kong Aik Lee. 2019. ASVspoof 2019: Future Horizons in Spoofed and Fake Audio Detection. arXiv:1904.05441. https://arxiv.org/ pdf/1904.05441.pdf [64] Ruben Tolosana, Ruben Vera-Rodriguez, Julian Fierrez, Aythami Morales, ary Javier Ortega-Garcia. 2020. Deepfakes sy ny ambiny: Fanadihadiana momba ny fanitsiana ny endrika sy ny fametrahana lainga. 64 (2020), 131–148. Information Fusion https://doi.org/10.1016/j.inffus.2020.06.014 [65] Xin Tong, Luona Wang, Xiaoqin Pan, ary Jingya Wang. 2020. An Overview of Deepfake: The Sword of Damocles amin'ny AI. . IEEE, 265–273. Fifanarahana iraisam-pirenena momba ny fahitana, ny sary ary ny fianarana maoderina amin'ny 2020 https://doi.org/10.1109/CVIDL51233.2020.00-88 [66] Luisa Verdoliva. 2020 Media Forensics sy DeepFakes: Fandaharana. 14, 5 (2020) ary 910–932. IEEE Journal of Selected Topics amin'ny famoahana famantarana https://doi.org/10.1109/JSTSP.2020.3002101 [67] Xin Wang, Junichi Yamagishi, Massimiliano Todisco, H ́ector Delgado, Andreas Nautsch, Nicholas Evans, Md Sahidullah, Ville Vestman, Tomi Kinnunen, Kong Aik Lee, Lauri Juvela, Paavo Alku, Yu-Huai Peng, Hsin-Te Hwang, Yu Tsao, Hsin-Min Wang, S ́ebastien Le Maguer, Markus Becker, Fergus Henderson, Rob Clark, Yu Zhang, Quan Wang, Ye Jia, Kai Onuma, Koji Mushika, Takashi Kaneda, Yuan Jiang, Li-Juan Liu, Yi-Chiao Wu, Wen-Chin Huang, Tomoki Toda, Kou Tanaka, Hirazu Kameoka, Ingmar Steiner, Driss Matrouf, Jean-Francois Bonastre 64 (2020), 27 pejy ao amin'ny pejy. Ny fiteny sy ny fiteny amin'ny solosaina https://doi.org/10.1016/j.csl.2020.101114 [68] Mirjam Wester, Zhizheng Wu, and Junichi Yamagishi. 2016. Analysis of the Voice Conversion Challenge 2016 Evaluation Results. In Ny fikambanana iraisam-pirenena momba ny fifandraisana amin'ny teny, 1637–1641. Ny fandaharam-potoana amin'ny Fivoriamben'ny Interspeech 2016 https://doi.org/10.21437/Interspeech.201 6-1331 [69] Junichi Yamagishi, Massimiliano Todisco, Md Sahidullah, H ́ector Delgado, Xin Wang, Nicholas Evans, Tomi Kinnunen, Kong Aik Lee, Ville Vestman, ary Andreas Nautsch. 2019. ASVspoof 2019: Automatic Speaker Verification Spoofing sy Countermeasures Challenge Evaluation Plan. https://www.asvspoof.org/asvspoof2019/asvspoof2019 evaluation plan.pdf [70] Zhao Yi, Wen-Chin Huang, Xiaohai Tian, Junichi Yamagishi, Rohan Kumar Das, Tomi Kinnunen, Zhen-Hua Ling, and Tomoki Toda. 2020. Voice Conversion Challenge 2020 – Intra-lingual Semi-parallel and Cross-lingual Voice Conversion –. In . International Speech Communication Association, 80–98. Ny fandaharam-potoana ho an'ny fanamby Blizzard sy ny fanamby amin'ny fifanakalozana feo 2020 https://doi.org/10.21437/VCC BC.2020-14 [71] Mohammed A. Younus sy Taha M. Hasan. 2020. Ny fomba fijery fohy amin'ny teknika fanadihadiana ny lahatsary. IEEE ny 115-120. Ny fandaharam-potoana amin'ny fihaonambe iraisam-pirenena faha-6 amin'ny taona 2020 https://doi.org/10.1109/IEC49899.2020.9122916 [72] Guangtao Zhai sy Xiongkuo Min. 2020 Ny fanombanana ny kalitaon'ny sary. 63, lahatsoratra 211301 (2020), 52 pejy. Ny siansa momba ny fampahalalam-baovao ao Shina https://doi.org/10.1007/s11432-019-2757-1 [73] Teng Zhang, Lirui Deng, Liang Zhang, ary Xianglei Dang. 2020. Ny fampianarana lalina amin'ny endriky ny maso: Fanadihadiana momba ny Deepfakes. ao amin'ny Izany dia mihoatra ny 67-70. Ny fandaharam-potoana amin'ny 2020 IEEE 3rd International Conference on Computer and Communication Engineering Technology https://doi.org/10.1109/CCET5090 1.2020.9213159 ao amin'ny firenena [74] Yuxuan Zhang, Huan Ling, Jun Gao, Kangxue Yin, Jean-Francois Lafleche, Adela Barriuso, Antonio Torralba, and Sanja Fidler. 2021. DatasetGAN: Efficient Labeled Data Factory with Minimal Human Effort. In IEEE, 10140 hatramin'ny 10150. Ny fifanarahana IEEE / CVF 2021 momba ny fahitana ny solosaina sy ny famantarana ny endrika https://doi.org/10.1109/CVPR46437.2021.01001 [75] ![](file:///C:/Users/user/AppData/Local/Temp/msohtmlclip1/01/clip_image051.gif)Yuanhan Zhang, ZhenFei Yin, Yidong Li, Guojun Yin, Junjie Yan, Jing Shao, and Ziwei Liu. 2020. CelebA-Spoof: Large-Scale Face Anti-spoofing Dataset with Rich Annotations. In . Springer, 70–85. Proceedings of the 2020 European Conference on Computer Vision https://doi.org/10.1007/97 8-3-030-58610-2 5 [76] Yuanhan Zhang, Zhenfei Yin, Jing Shao, Ziwei Liu, Shuo Yang, Yuanjun Xiong, Wei Xia, Yan Xu, Man Luo, Jian Liu, Jianshu Li, Zhijun Chen, Mingyu Guo, Hui Li, Junfu Liu, Pengfei Gao, Tianqi Hong, Hao Han, Shijie Liu, Xinhua Chen, Di Qiu, Cheng Zhen, Dashuang Liang, Yufeng Jin, ary Zhanlong Hao. https://arxiv.org/pdf/2102.12642.pdf [77] Bo Zhao, Shaozeng Zhang, Chunxue Xu, Yifan Sun, and Chengbin Deng. 2021. Deep Fake Ge-ography? When Geospatial Data Encounter Artificial Intelligence. 48 amin'ny 4 (2021), 338 hatramin'ny 352 Cartography sy ny siansa momba ny fampahalalana ara-jeografika https://doi.org/10.1080/15230406.2021.1910075 [78] Peng Zhou, Xintong Han, Vlad I. Morariu, ary Larry S. Davis. 2017. Ny tambajotran'ny neuron roa ho an'ny famantarana ny endriky ny maso. . IEEE, 1831–1839. Ny fandaharam-potoana amin'ny 2017 IEEE Conference on Computer Vision and Pattern Recognition https://doi.org/10.1109/CVPRW.2017.229 [79] Tianfei Zhou, Wenguan Wang, Zhiyuan Liang, ary Jianbing Shen. 2021. Face Forensics in the Wild. . IEEE, 5774–5784. Ny fifanarahana IEEE / CVF 2021 momba ny fahitana ny solosaina sy ny famantarana ny endrika https://doi.org/10.1109/CVPR46437.2021.00572 [80] Jun-Yan Zhu, Taesung Park, Phillip Isola, ary Alexei A. Efros. 2017. Tsy mifanaraka amin'ny sary amin'ny sary mampiasa tambajotra mifanohitra amin'ny tsingerintaona. IEEE, 2242 hatramin'ny 2251 Proceedings of the 2017 IEEE International Conference on Computer Vision https://doi.org/10.1109/ICCV.2 017.244 [81] Bojia Zi, Minghao Chang, Jingjing Chen, Xingjun Ma, and Yu-Gang Jiang. 2020. WildDeepfake: A Challenging Real-World Dataset for Deepfake Detection. In . ACM, 2382–2390. Proceedings of the 2020 28th ACM International Conference on Multimedia https://doi.org/10.1145/339417 1.3413769 Ity lahatsoratra ity dia azo jerena ao amin'ny archiv amin'ny lisansa CC by 4.0 Deed (Attribution 4.0 International). Ity lahatsoratra ity dia azo jerena ao amin'ny archiv amin'ny lisansa CC by 4.0 Deed (Attribution 4.0 International).