څنګه څیړونکو د AI تبادلې اندازه کولو، تشخیص کولو او بنچ کولو لپاره

د نویسنده: 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) د نویسنده: Enes Altuncu، ea483@kent.ac.uk (د کانت پوهنتون، UK) ویرجینیا N. L. Franqueira، V.Franqueira@kent.ac.uk (د کانت پوهنتون، UK) Shujun Li، S.J.Li@kent.ac.uk (د کانت پوهنتون، UK) Abstract د Abstract د AI، په ځانګړې توګه عمیق زده کړې کې وروستیو پرمختګونه، د نوي واقعي ښکلي مصنوعي رسنیو (ویډیو، انځور، او آډیو) او د موجودو رسنیو کنټرول په جوړولو کې د پیاوړتیا زیات کړي، کوم چې د "د عمیق فاکس" نوي اصطلاح جوړولو ته اړتیا لري. په انګلیسي او چين کې د څیړنې کتابتون او سرچینو پر بنسټ، دا کاغذ د عمیق فاکس په پراخه کچه ښیښه کوي، د دې پیاوړي مفهوم د ډیرو مهمو اړخونو پوښښ کوي، په شمول 1) مختلفو تعریفونه، 2) عموما کارول شوي کړنو معیارونه او معیارونه، او 3) د عمیق فاکس سره تړل شوي ډاټا سیټونه، چټکونه، : Deepfake، Survey، Definitions، Datasets، Benchmarks، Challenges، Competitions، Standards، Performance Metrics. د Keywords 1 نندارتون د AI او ماشین زده کړې په وروستيو کې پرمختګونه د ویډیو، انځور، او آډیو په توګه ډیر واقعي رسنیو تولید کولو وړتیا زیات کړي. په ځانګړې توګه، د پرمختللي عمیق زده کړې روشونه د "د پوښونو" تولید ته امکان ورکړي، د کنټرول شوي یا مصنوعي رسنیو چې په حقیقت کې د انسان په چټکۍ سره ناڅاپي شتون نلري. که څه هم deepfake د نسبتا نوي نښې ده (د 2017 په پای کې لومړی ښودل شوی)، د هغې د ودې حیرانتیا لري. د 2019 او 2020 Deeptrace راپورونو سره د deepfake حالت په اړه [ په انګلستان کې د انټرنټ کې د deepfake ویډیو شمېر له 7964 (د 2018 دسمبر) څخه 14 678 (د 2019 جولای) ته 85 047 (د 2020 دسمبر) ته وده ورکړي، چې د 2018 څخه د 2020 په پرتله 968٪ وده ورکړي. 2 په دې کار کې، موږ د موجوده deepfake-related څیړنې اکوسیستم په مختلفو اړخونو کې څیړنه کوو، په شمول د فعالیت د نمونې او معیارونو، ډاټا سیټونه، چټکونه، رقابتیونه او بیلګې په توګه. برسېره پر دې، موږ د 12 انتخاب شوي deepfake-related څیړنې کاغذونو د میټا بیاکتنې وړاندې کوو چې په سیستماتیک ډول د ذکر څخه غیر چند اضافي اړخونو کې شامل دي، لکه د فعالیت مقایسه، د مهمو چټکونو او سپارښتنې. په دې سروې کې، موږ د نسبتا ډیر پراخ روښانه لارښوونې لرو چې په پراخه کچه د deepfakes په پام کې دي. موږ هم د بیومیمریک او ډیریډیمیا فیکس په شمول تړاو لري، لکه څنګه چې deepfakes معمولا د بیومیمریک پر بنسټ د تصدیق سیسټمونو په اړه د وړاندې کولو حملاتو پیل کولو لپاره کارول کیږي او د deepfakes کشف د ډیریډیا فیکس په برخه کې وي. د "deepfake" د مختلفو تعریفونو په اړه نور تفصيلات بحث دی وروسته. 1.1 د اصطلاح Deepfake تعریفونه لکه څنګه چې خپل نوم ښودل کیږي، د "deepfake" اصطلاح د "deep" (د اشاره کولو لپاره) ترکیب څخه رامینځته کیږي. (DL)) او "د جعلي". دا په عمومي توګه د موجودو رسنیو (صور، ویډیو او / یا غږ) یا د نوي (سټیټیک) رسنیو تولید لپاره د DL پر بنسټ طریقو کارولو لپاره کارول کیږي. په عمومي توګه بحث شوي deepfake معلوماتو د جعلي چڼاسکه، جعلي کلام جعلي، او جعلي ویډیو چې د جعلي انځورونه او جعلي کلام جعلي ترکیب کوي. په داسې حال کې چې د کلمې کې "د جعلي" د manipulated یا synthesized رسنیو لخوا اشاره کوي، د deepfake تکنالوژۍ ډیری بیلابیلو غوښتنلیکونه شتون لري، د تفریح او تخنیکي هنر لپاره. د دې په اړه، یو بل اصطلاح "د جعلي synthesis" په توګه د غیرقانوني غږ بدلون په توګه پیشنهاد ]. د دې نوی اصطلاح، خو په پراخه کچه تصویب نه شو. د عالي زده کړې 60 لکه څنګه چې په دې وروستيو کې ذکر شوي، د عالي فاکسونو او غیر عالي فاکسونو تر منځ د مرز په واقعیت کې یو واضح قطع نه ده. موږ باید لږ تر لږه د دوو مهمو اړخونو په پام کې ونیسئ، یو د عالي فاکسونو کشف او بل د جوړولو په اړه. لومړی، د Deepfakes تشخیص اغیزمنې په څیر د DL تکنالوژۍ په کارولو سره تولید شوي روښانه فایبرونو د تشخیص په څیر ورته روښانه لارښوونه پیژندل کیږي. پرمختللي فایبرونه هم د DL کارولو ته وده ورکړي ترڅو دوی د فعالیت د ښه کولو لپاره وده ورکړي، مګر دوی اړتیا نلري چې څنګه د سایټ رسنیو جوړ شوی دی (د عمیق یا نه). په ځینې کچه، یو کولی شي argue چې د Deepfakes تشخیص د Deepfakes ځانګړي روشونو پراختیا نه لري (چې ځینې څېړونکو دا غوره کوي)، مګر یو قوي او جامع detector چې هر (د عمیق یا نه) فایبر رسنیو سره کار کوي. دا د دوو اړوند موضوعاتو لپاره لیدل کیدی شي: دوهم، تاسو کولی شئ argue that the word “deep” in “deepfake” does not necessarily refer to the use of “deep learning”, but any “deep” (i.e., sophisticated) technology that creates a very credible fake media. ] د Deepfake په توګه د آډیو ویزايي تبادلې په توګه د "د تخنیکي تخنیکي ... او تکنیکونو لړ" په کارولو سره په پام کې ونیسئ. او ، د "د (سانه) لاسرسي سافټویر سره (یا هیڅ سافټویر سره) جوړ شوي آډیو ویډیو میډیاونو د ټیټ کچه manipulation" ته اشاره کوي چې د موادو په چټکۍ سره، چټکولو، ریستګ یا re-contextualizing کړي. د "deepfake" په دې پراخه تفاهم کې هم د قانون جوړونکو لخوا د بدیل deepfakes مبارزه لپاره د نوي قانونونو لپاره تصویب شوي دي. د مثال په توګه، د لاندې دوو متحده ایاالتو قانونونه د "deepfakes" په څیر تعریف کوي: 9 د فحش د فحش د 2018 کال د بدعتو عمده جعلي ممنوعیت قانون 1: 1041 (b) (2) : « ” د "د عمیق جعلي" اصطلاح په دې توګه جوړ شوي یا بدل شوي آډیو ویډیو ویډیو ریکارډ معنی لري چې دا ریکارډ به په غلطي توګه د منطقي تماشایونکي لپاره د یو فرد په واقعي خبرې یا عمل کې د حقیقي ریکارډ په توګه ښکاري. 2019 د عمیق فاکس د حسابولو قانون 2: §1041.(n) (3) : « د "د عمیق فحش" اصطلاح د ویډیو ریکارډ، فلم فلم، غږ ریکارډ، برېښنايي انځور، یا انځور، او یا هر تکنالوژیکي انځور د خبرې یا عمل په عمده توګه له دې څخه رامینځته کیږي. (A) چې په واقعیت کې د هغه شخص د هر کلام یا عمل چې په واقعیت کې د دې کلام یا عمل نه کوي، په واقعیت کې ښودل کوي؛ او ” (B) چې د تولید په عمده توګه د تخنیکي وسایلو څخه پرته د نورو شخصونو وړتیا څخه د دې شخص په فیزیکی او یا کلمې توګه وکاروي. لکه څنګه چې موږ د "deepfake" په دې قانوني تعریفونو څخه وګورئ، د DL په توګه د تکنالوژۍ په توګه د کارولو په بشپړه توګه ندي. دلته د "اټیټیټیت"، "پایه" او (په هر ډول کې) "تقنيات" تمرکز کوي. 1.2 کچه او د انعام د deepfake تعریفونو په اړه د اوسط بحث په اساس، موږ کولی شو وګورئ چې د deepfake څخه د غیر عمیق فاکس څخه مختلفولو تل په مستقیم ډول یا معنوي نلري. برسېره پر دې، زموږ په کارولو ارزښت او مقایسه کې زموږ تمرکز لپاره، د deepfake او غیر عمیق فاکس تر منځ د محدودیتونه ډیر مخکښ دي. دا ځکه چې DL یوازې د ماشین زده کړې (ML) ځانګړي (د عمیق) ډول دی، او په پایله کې، DL او غیر عمیق ML روشونه ډیری عام مفهومونه، مټريکونه او پروسهونه شریک کوي. که څه هم چې deepfake کولای شي په ډیر پراخه معنی کې درک شي، په دې کار کې، موږ د ډیری موضوعاتو پوښښ څخه مخنیوی لپاره خورا مخنیوی تمرکز لري. موږ، له دې امله، د دې سروې په پراخه کچه د دې څېړنې په څیر تعریف کړي: د نمونې او معیارونو لپاره، موږ غوره کړ چې د عمومي ML روشونو ارزونې لپاره په عمومي ډول کارول کیږي او د deepfake د جوړولو یا د څارنې روشونو ارزونې لپاره په ځانګړې توګه تعریف شوي. د ډاټا سیټونه، چټکونه، رقابتیونه او بیلګې په توګه، موږ د ډیرو فاکسونو سره تړاو لري چې د سروې کاغذونو کې شامل دي او هغه چې د "د ډیرو فاکس" یا ورته اصطلاح په ځانګړې توګه اشاره کوي. For the meta-review, we considered only survey papers whose authors explicitly referred to the term “deepfakes” in the meta data (title, abstract and keywords). 2 د نمونوي په دې سروې کې پوښل شوي څیړنې کاغذونه (یا د deepfake-related سروې کاغذونه) د علمي ډاټا بیسونو، Scopus او چین آنلاین ژورنالونو (COJ) په سیستماتیک ډول څیړنې له لارې مشخص شوي دي. په Scopus او COJ کې د څیړنې لپاره د لاندې څیړنې پوښتنو کارول: 3 (deepfake* یا deep-fake* یا "deep fake*") او (تقييم يا سروې يا په څیر يا systemati* یا SoK) (deepfake OR 深度伪造) AND (综述 OR 进展) د څیړنو په انګلیسي کې 41 سروې کاغذونه او په چين کې 15 سروې کاغذونه راټول شوي. د دغو کاغذونو څخه، د انګلیسي کې چاپ شوي 8 او د چين کې چاپ شوي 4 کاغذونه راټول شوي دي. Deepfake-related challenges, competitions and benchmarks were identified via multiple sources: the survey papers selected, research papers from the co-authors’ personal collections, Google Web searches, and manual inspection of websites of major AI-related conferences held in 2020 and 2021 (where such challenges and competitions are routinely organised). The inspected conferences include those listed in the ACL (Association for Computational Linguistics) Anthology ، ICCV، CVPR، AAAI، ICML، ICLR، KDD، SIGIR، WWW، او ډیری نور. برسېره پر دې، د ډاټا سیټونو بشپړ لیست د انتخاب شوي سروې کاغذونو او مشخص شوي چټکونه، رقابتیونه او بیلګې په اساس جوړ شوی. اړین معیارونه په عمده توګه په دې سروې کې شامل شوي څیړنې کاغذونو، د همکارانو شخصي معلوماتو، او د Google ویب څیړنې له لارې مشخص شوي. د فعالیت د معیارونو لپاره، موږ د اړونده معیارونو، د سروې کاغذونو، او مشخص شوي چټکونه، رقابتیونه او بیلګې په اساس کارول کیږي. 4 3 Deepfake-related کړنو معیارونه او معیارونه په دې سروې کې، موږ د فعالیتونو د ارزیابی او د deepfake توليد او د تشخیص لارښوونې په پرتله تمرکز کوي. د دې فعالیتونو د ارزیابی لپاره کارول شوي مټريکونه زموږ د بحثونو په کور کې دي. په دې برخه کې، موږ د فعالیتونو مټريکونه چې عموما د deepfake توليد او د تشخیص algorithms په ارزښتونو کې کارول کیږي په لټه کې دي. یاد ولرئ چې ټول مټريکونه چې په دې برخه کې پوښښ شوي دي هم عموما د ورته سیستمونو د فعالیتونو د ارزیابی لپاره کارول کیږي چې د deepfake توليد یا د تشخیص لپاره نه دي. له همدې امله، دا برخه د عمومي فعالیت مټريکونو په اړه یو خورا مختصر لارښوونې په توګه لیدل کیږي. په وروستۍ برخه کې، موږ هم په خلاصه توګه بحث کوو چې څنګه د اړوند کړنو معیارونه په رسمي معیارونو کې پوښل کیږي. د "د رسمي معیارونو" له لارې، موږ د معیارونو ته اشاره کوو چې د رسمي پروسه له مخې تعریف شوي دي، معمولا د یو یا د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیارونو د معیار د نړيوال electrotechnical کمیسیون (IEC) په یاد ولرئ چې موږ په نظر کې دي چې د سندونو په پراخه کچه د معياري سازمانونو، د بیلګې په توګه، International Telecommunication Union (ITU) د معياري په توګه تعریف شوي دي. د ISO تخنیکي راپورونه (TRs) 5 6 7 4.1 د اندیښنې مټریکس Deepfake تشخیص په عمده توګه د باینری کټګوریشن ستونزه ده. د باینری کټګوریټر د انټرنټ ترلاسه کوي چې یا او د يو باینری ارزښت له لاسه ورکوي چې دا یا د مثال په توګه، د deepfake تشخیص سیسټم به د شواهدې انځور په توګه د انټرنېټ په توګه واخلئ چې ممکن یا او د محصول یا . په حقیقت کې مثبت په حقیقت کې د منفي مثبت پیژندل منفي پیژندل په حقیقت کې د جعلي په حقیقت کې مخکښ مخکښ د واقعي پیژندل د Binary Classifiers ارزښتولو لپاره کارول شوي یو بنسټیز وسیله دا دی چې چې د ټولګي ماډل د بریالیتوب او ناکامۍ په اړه خلاصه کوي. په یوه لړ کې د دوو ارزښتونه او په بل لړ کې د دوو ارزښتونه. ټولګي ده (د حقیقي مثبت او حقیقي منفي) کله چې د واقعي او مخکښ ارزښتونه یوځای کیږي. (د فحش مثبت او فحش منفي) کله چې د واقعي او توقع شوي ارزښتونه برابر نه وي. جدول د دوامداره deepfake کټګوریو لپاره د اندیښنې مټریکس ښيي. د دوو سیلونه په سبز کې، TP (د ) او TN (د شمېر ), indicate correct prediction results, and the two cells in red, FN (the number of ) او FP (د تعداد ) ، د غلطو په دوو ډولونو کې د غلطو پیژندنې پايلې ترسره کولو په وخت کې اشاره کوي. confusion matrix اوسني پیژندل بریالیتوب / درست / حق غلط / غلط / غلط 1 true positives true negatives false negatives false positives جدول 1: د deepfake کشف لپاره د باینری کټګوریو لپاره د اندیښنې مټریکس. fake (predicted) real (predicted) fake (actual) TP FN real (actual) FP TN جعلي (د واقعي) د TP د متحده ایالاتو واقعي (د واقعي) د FP د 3.2 د دقت او یادښت پر بنسټ په برخه کې داخل شوي چار بنسټیز ارزښتونه د TP، TN، FP او FN، موږ د دوو مهمو دنده مټريکونه د باینری klasifier لپاره تعریف کوو - او . 3.1 precision recall د باینری کټګوریټر د دقت په توګه د فریکونسۍ د د نمونې په منځ کې ټول . په نښلیدو مټریکس کې، دا د اصل نمونې درملنه ده. دا کولی شي په رسمي توګه د Eq په توګه تعریف شي. ( همدارنګه په حقیقت کې مثبت مثبت پیژندل 1 کله چې د مثبت او منفي نمونې تر منځ د "لري" نسبت د ازموینې ټولګه څخه ډیر مختلف دی، دا اغیزمنه ده چې د غلط مثبتو وزن تنظیم کړئ، کوم چې له دې امله د (د WP) په Eq. ( ، چې 0 د منفي او مثبت نمونې ترمنځ د نسبت له خوا د وزن دی. weighted precision 2 د A د Binary Classifiers یادونه ده د د نمونې په منځ کې د نمونې، لکه څنګه چې په Eq. ( ). په نندارتون کې، دا په لومړي لړ کې د حق نمونې فریکونس دی. مثبت پیژندل په حقیقت کې مثبت 3 د مثال په توګه، د باینری کټګوریو په اړه فکر وکړئ چې د معلوماتو د ډاټا څخه د انځور په اړه چې د deepfake او حقیقي (خود) انځورونه لري یا نه دي یاست. د کټګوریو د دقت د ټولو د deepfake په توګه کټګوریو شوي انځورونو په منځ کې د درست ډول کټګوریو شوي انځورونو برخې ده. په بل کې، recall د deepfake انځورونو برخې دی چې د کټګوریو له لارې په ډاټا کې د ټولو deepfake انځورونو کې تشخیص شوي دي. 3.3 حق او غلط مثبت نرخونه د پیژندل شوي مثبت نمونې په تمرکز کې، موږ کولی شو د دوو نمونې هم تعریف کړو: (TPR) ، هم د نوم (CDR) په توګه د واقعي مثبت نمونې او (FPR) ، هم د نوم (FAR)، لکه څنګه چې په Eqs کې ښيي چې په واقعیت کې د منفي نمونې ترمنځ د پیژندل شوي مثبت نمونې برخې. ( ) او true positive rate correct detection rate false positive rate false alarm rate 4 ( ). په اندیښنې مټریکس کې، TPR د لومړي لړ کې د پیژندل شوي مثبت نمونې فریکونس دی او FPR د دویم لړ کې د پیژندل شوي مثبت نمونې فریکونس دی. یاد ولرئ چې TPR په اصل کې د (د نندارتون کې) ) ) . 5 recall 3 3.4 حق او جعلي منفي نرخونه لکه څنګه چې د حق او غلط مثبت نرخونه، موږ کولی شو د دوو نورو نرخونه د منفي پیشې پایلې تمرکز کوي: (TNR) چې د واقعي منفي نمونې په منځ کې د پیژندل شوي منفي نمونې فریکونسۍ ندي، او (FNR) په Eqs کې ښيي چې په واقعیت کې مثبت نمونې کې د پیژندل شوي منفي نمونې فریکونسۍ ښيي. ( ) او ( همدارنګه true negative rate false negative rate 6 7 3.5 حساسیت او ځانګړتیا په ځینو غوښتنلیکونو کې، په ځانګړې توګه په بيولوژي او درملنې کې، د TPR او TNR ډیر عام کارول کیږي، او دوی اغیزمن دي. (د TPR) او (TNR). د دې دوو اصطلاحاتو تمرکز د مخکښ پایلو دوو ډولونو په اړه ده. دا په ډیرو فاکس سره تړاو څیړنو کې لږ کارول کیږي، له همدې امله موږ به په دې کاغذ کې د نورو په اړه اشاره نه کوو. sensitivity specificity 3.6 د مساوي غلطی کچه د غلطی نرخونو په اړه تمرکز کول معنی لري چې موږ باید د FPR او FNR په پام کې ونیسئ. دا دوه نرخونه په معمول ډول د یو نرخ کمولو په معمول ډول د نورو نرخونو زیاتولو ته وده ورکوي. له همدې امله، د هر دوو غلطی نرخونو په ورته وخت کې کمولو هڅه کولو لپاره، کوم چې په معمول ډول غیر ممکن دی، د عملی غوښتنلیکونو کې ډیر واقعي کار دا ده چې د مناسب توازن ترلاسه کړي نو دوی په هر دوو کې د معافیت وړ حد لاندې وي. په ځینو غوښتنلیکونو کې، لکه بیومیمریک، خلک په ځانګړي ډول د دې چې د دې لپاره چې د دې په توګه د "د" د جوړولو لپاره اړتیا لري. او یا د EER / CER د ځینې غوښتنلیکونو لپاره اړتیا نلري، په ځانګړي ډول کله چې د دوو ډولونه غلطۍ د مختلفو اړخیزو کچه دي، د مثال په توګه، د مهمو deepfakes (د بیلګې په توګه، فلیټ خبرونه چې کولی شي د هغه څه چې خلک خپل وو ته اغیزه کوي) څومره موږ کولی شو فلیټ مثبتونه (د فلیټ تفتیشونه) د فلیټ منفيونه (د لګولو تفتیشونه) څخه ډیر tolerate. equal error rate crossover error rate 3.7 د دقت او F-Score د EER / CER په پرتله، د نورو مترجمونه هم شتون لري چې هڅه کوي چې د دوو ډولونه غلطاتو ته وده ورکړي، ترڅو د بیلابیلو کټګورۍ په ټولیزه فعالیتونو کې ډیر توازن ترلاسه کړي. د دوو په عمده توګه کارول شوي دي او (د دې نوم هم ). د دوو میټریکونه کولی شي د چار بنسټیز ارزښتونو (TP، TN، FP، او FN) په اساس تعریف شي. accuracy F-score F-measure د Binary Classifiers د دقت په توګه د فریکونسۍ د د نمونې (د حقیقي مثبت او حقیقي منفي) د مجموعي نمونې په شمول د نمونې په شمول د Eq. ( همدارنګه په درست ډول پیژني 8 د باینری klasifier F-score په واقعیت کې د مترو د کورنۍ ده. دا عمومي شکل کولی شي د پارامتر په اساس بیان شي لکه څنګه چې په Eq. ( همدارنګه β 9 د ټولو F-کاسټونو په پراخه کچه کارول شوي نسخه دا دی چې د ، چې په واقعیت کې د F-score سره = 1. په ځانګړې توګه، دا په Eq. ( همدارنګه F1-score β 10 3.8 د راټولر د کارولو ځانګړتیاوې کرښه او د کرښه لاندې سيمه (ROC) curves په عمده توګه د باینری klasifiers چې د پیژندنې د پايلې (یا احتمال) له لاسه اندازه کولو لپاره کارول کیږي. Receiver operating characteristic Consider the following. Let be the set of all test samples and let the output scores ( ) (for all ∈ ) lie in the interval [ ] on the real line. Let ∈ [ ] be a prediction threshold for the model, and assume that the classifiers works as follows for all ∈ : S f s s S a, b t a, b s S It is easy to see that, for = د , all the samples will be classified as positive, leading to FN = TN = 0 so TPR = FPR = 1; while for = ، ټول نمونې به د منفي په توګه راټول شي، د FP = TP = 0 له امله د TPR = FPR = 0 له امله. and ، د TPR او FPR ارزښتونه به په عادي ډول د 0 او 1 تر منځ وي. 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 د random classifier لپاره، په فرض کې چې دا ( ) 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 Another widely used performance metric for binary classifiers that can return a probability score for the predicted label is د حق لیبل سره د باینری طبقه کولو لپاره ∈ {0*,* 1} او د اندازې احتمال د پروپیلن ( = 1)، د نندارتون لګښت د نندارتون لګښت احتمال دی چې د کثافاتو لپاره د سچا نندارتون ورکړل کیږي، لکه څنګه چې په 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. ( ، چې is 1 if the -th sample is true and 0 if false, and ˆ is the predicted probability of = 1. n 13 yi i y i د 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), which has been widely used by the signal processing and multimedia communication communities, including digital TV and other multimedia-related consumer applications. As its name implies, MOS is calculated by averaging the subjective scores given by a number of human judges, normally following a numerical scale between 1 and 5 or between 0 and 100. MOS has been used in some deepfake-related challenges (see Section ) او همدارنګه د Deepfake ډاټا سیټونو د کیفیت (د واقعیت / طبيعت) تبادله کولو لپاره (د برخه وګورئ) ). mean opinion score 5.2 4.6 As a general subjective PQA metric, MOS has been standardised by the ITU همدارنګه د ITU معیارونه لري چې د ویډیو کيفيت ارزیابی (VQA) معیارونه او د VQA کاروونکي مطالعې ترسره کولو لپاره معياري پروسهونه مشخصوي، د مثال په توګه، د ITU-T سپارښتنه P.910 "د multimedial غوښتنلیکونو لپاره د ویډیو کيفيت ارزیابی معیارونه" . 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 د ځانګړي PQA معلوماتو د کنترول په توګه د MOS په څیر یو لوی ستونزه ده چې د انسانو د قضاوتونو کارولو او د یو ښه کنترول فیزیکی ازمايښت چاپیریال او پروتوکول لري، چې د ډیری غوښتنلیکونو لپاره آسانه نه دي. د PQA سره تړاو د کاروونکو مطالعې ترسره کولو هڅو او لګښتونو کمولو ته مرسته وکړي، د مختلفو objektive PQA مټريکونو په عمده توګه په درې کټګوریو کې تقسیم شوي دي: بشپړ رینج (FR) مټريکونه (که د دې واقعیت ته اشاره کوي چې د دې اصل "نمايښت کیفیت" سیگنال د رینج په توګه شتون لري)، بشپړ رینج (RR) مټريکونو په اتوماتيک ډول مح 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., (د 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 , the joint technical committee for standardising “information technology”. 5 10 11 د ISO / IEC JTC / SC 42 یو وروستۍ هڅه ده چې د ISO / IEC TR 24029-1:2021 "د مصنوعي انټرنټ (AI) - د عصري شبکې د قوي کیفیت تقييم - برخه 1: د مجموعې" تولید کړي. , a technical report (TR) that systematically covers many commonly used performance assessment concepts, methods and metrics. Although the technical report has “neural networks” in its title, most performance assessment concepts, methods and metrics included are common ones for all supervised machine learning models. 12 د فعالیت د نمونې په اړه، د ISO / IEC JTC 1 / SC 42 د دوو نورو اوسني کار برخو چې د توجه وړ دي لکه څنګه: ISO/IEC DTS (Draft Technical Specification) 4213 “Information technology – Artificial Intelligence – Assessment of machine learning classification performance” 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)، کوم چې د معياري FPR او FNR ته اشاره کوي. دا معياري هم د واحد شمیره متريونو د کارولو په ګډون د EER او AUC (که په پراخه کچه په بیومیمريک په اړه د څیړنې په پراخه کچه کارول شوي دي) په کارولو څخه مخنیوي. 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) میکانیزمونه، کله چې د اصطلاح " ” په دې کې د ” ”. 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 کله چې deepfake انځورونه یا ویډیوونه د بیومیمريک سیسټمونو د جعليولو لپاره کارول کیږي، د دې معياري معياري به مهم وي. 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 Deepfake-Related Datasets په دې برخه کې، موږ د deepfake-related ډاټا سیټونه چې موږ د deepfake-related سروې کاغذونو د میټا-تقييم څخه مشخص شوي دي، د deepfake-related چټکونو، رقابتیونه او بینچینکونه پوښښ، په GitHub کې د deepfake-related ډاټا سیټونو یو آنلاین ټولګه , 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 [ ]: دا ډاټا سیټ د 131،500 چیرې انځورونه لري، په ګډون 78،500 انځورونه د FaceForensics ډاټا سیټ کې د 150 ویډیو څخه اخلي او 53،000 انځورونه د یوټیوب څخه د 150 جعلي ویډیو څخه اخلي. Fake Faces in the Wild (FFW) dataset 32 : دا د سوداګرۍ ډاټا سیټونه دي چې د Generated Media، Inc. لخوا وړاندې شوي دي، سره د StyleGAN لخوا تولید شوي تقریبا 2.7 میلیونه مصنوعي چیرې انځورونه. د 10،000 128x128 مصنوعي انځورونو سره وړیا نسخه د علمي څیړنې لپاره شتون لري. د ویب پاڼه هم د تعاملاتي چیرې جنراتور وړاندې کوي. 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 [ ]: This dataset includes 100,000 synthesised face, bedroom, car and cat images by a GAN generator trained based on real images in the FFHQ and LSUN datasets (three object types – bedrooms, cars and cats – for the latter). Note that the name “100K-Generated-Images” was not a proper one as the authors [ ] 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 [ ]: 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 [ ]: This dataset includes 40,000 images, half real and half deepfake. The images were collected from four sources: the CelebA-HQ dataset , the Flickr-Faces-HQ dataset , the 100K-Faces dataset (د generated.photos ډاټا سیټونو په تفصیل کې وګورئ) او . Faces-HQ 21 27 28 29 thisperson-doesnotexist.com [ ]: This dataset includes 625,537 synthesised face images of 10,177 celebrities, with 43 rich attributes on face, illumination, environment and spoof types. The real images were selected from the CelebA dataset . 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 او د 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 [ ]: دا ډاټا سیټ 98 فایبر ویډیوونه لري، د نږدې نیم (49) د یوټیوب څخه ډاونلوډ شوي حقیقي ویډیوونه دي، او د بل نیم د FakeApp ګرځنده غوښتنلیک په کارولو سره تولید شوي فایبر ویډیوونه دي (که اوس بند شوی دی). د ویډیو ډاټا سیټ د عمیق فایبر ویډیو څیړنې په اساس د چټک کولو رفتار څیړنې لپاره کارول کیږي، نو ټول ویډیوونه لږ تر لږه یو د چټک فایبر واقعیت لري. ټول فایبر ویډیوونه د اصلي چیرې په هر د حقیقي ویډیو کې د کارونکي Nicolas Cage په چیرې بدلولو سره جوړ شوي دي په دې توګه، یوازې یو موضوع نلري. 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 [ ]: This dataset contains 3,363 face videos, covering 28 subjects, gender, and skin colour. It was created as a joint effort between two units of Google, Inc.: Google AI 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 [ ]: 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 [نور] ]: 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 [ ]، د ریرسون آډیو ویزیو ډاټا بیس د احساساتي خبرې او آواز (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 [ ]. دا د 216 بیانات (162 لپاره روزنې او 54 لپاره ازمايښت) له 10 سپارلو څخه هر سپارلو لپاره شامل دي. ګډون کوونکي ټیمونه (17) د 25 سرچینې هدف سپارلو ګروپونو لپاره خپل VC سیسټمونه پراختیا، او بیا د ارزیابی لپاره تولید شوي بیانات وړاندې کړي. لږ تر لږه 6 ګډون کوونکي ټیمونه په خپل VC سیسټمونو کې DL-related تکنیکونه (LSTM، DNN) کارول (د پایلې تحلیل کاغذ جدول 2 وګورئ) ), so the submitted utterances can certainly be considered deepfakes. Voice Conversion Challenge 2016 database 62 47 39 [ ]: 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 [ ]: This dataset is based on the Effective Multilingual Interaction in Mobile Environments (EMIME) dataset , 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 [ ]. په یاد ولرئ چې د نندارتون سازمانونکي د "deepfake" اصطلاح په ځانګړې توګه نه کارول، په هرصورت چې د DL-generated لګولو لګولو معلوماتو کولی شي د deepfakes په پام کې ونیسئ. 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 ډاټا سیټونه : These datasets were created by the DARPA Media Forensics (MediFor) Program for the 2020 OpenMFC . د دوو GAN-generated deepfake ډاټا سیټونه شتون لري، یو چې د 1,000 څخه زيات deepfake انځورونه لري او د نورو سره د 100 څخه زيات deepfake ویډیوونه. د ډاټا سیټونه یوازې د رقابتی شرکتونو لپاره ثبت شوي دي. 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 at the 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 په دلچسپي کې، موږ کولی شو د مختلفو ډاټا سیټونو د واقعیت کچه ترمنځ یو لوی فرق وګورئ. هغه څه چې احتمالا ډیری حیرانتیا لري دا دی چې FaceForensics++، چې تر ټولو پراخه توګه کارول شوي deepfake ډاټا سیټونو څخه یو دی، د MOS کچه ډیر ټیټ دی او د 9٪ څخه کم ګډون کوونکي د 30 انتخاب شوي ویډیوونو لپاره حقیقي دي. 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 [ ]. د جنګ او ال. 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 ، په کوم کې د لومړي دوو ډاټا سیټونه (DeeperForensics-1 او Celeb-DF) د نوي نسل (2020) په برخه کې دي، په داسې حال کې چې نورو د قديم نسل څخه دي. د دوو نسلونو په ګډون، د نوي نسل په 2021 کې هم ښودل شوی: ډیری ډاټا سیټونه د ډاټا سیټونو په پرتله ډیر واقعي ډاټا سیټونو ته تمرکز کوي (د دې لپاره، په وحشي کې) یا د ډاټا سیټونو په پرتله ځانګړي سيمو کې (د مثال په توګه، د ډاټا سیټونه) 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 The Deepfake Detection Challenge (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 د ټولو ټانډیم سیسټمونو د کړنو په اړه ارزښت کوي چې په دې توګه " ”. It is calculated according to Eq. ( ), where cm ( ) and cm( ) are, respectively, “ د 54 خبرې DeepFake 55 56 13 EER reporting is deprecated 57 4 63 69 د CM [انټرنېټ] د 'پورت' په توګه کار کوي ترڅو مشخص کړي که آیا یو ډاټا خبرې input له bona fide (د حقیقي) کاروونکي څخه رامینځته کیږي، مخکې له دې چې دا د اصلي بیومیمریک verifier (د ASV سیسټم) ته ورسیږي. 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. ( ), including constants 1 and 2, please refer to the ASVspoof 2019 evaluation plan [ ننداره 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 process. The first phase is the “development phase”; it started in April 2021 when the 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 The FaceForensics Benchmark 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 . The outcome of a submission is illustrated in Figure , where the scores are a measure of accuracy (Eq. ( )). 65 66 61 56 4a 4b 8 د Open Media Forensics چټک (OpenMFC، د DARPA MFC مخکې) 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 SenseHuman Workshop . 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 is the number of videos in the hidden test set, is the ground truth label of video (Fake:1, real:0) او ( ) 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 The Face Forgery Analysis Challenge 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 0 0 ،* 0*.*95)، او د متوسط یادښت (AR) په (AR@ ) where د Top labels returned for multi-class classifiers. 74 75 25 t 5 K K K K د 2020 کال CelebA-Spoof Face Anti-Spoofing چمتو کول 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 , Faceswap-GAN, Recycle-GAN [ ] 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 د 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 . The Multimedia Information Recognition Technology Competition included two tasks on deepfakes: one on deepfake video detection and one on deepfake audio/speech detection . د Deepfake ویډیو تشخیص ورکشاپ د 3000 ویډیو کارولو لپاره کارول، او د لګولو کڅوړه په توګه یوازې کارول شوی. د Deepfake آډیو / کلام تشخیص ورکشاپ د 20،000 آډیو نمونې کارولو لپاره کارول شوی (د عموما په چين کې، او په انګلیسي کې) او د EER په توګه یوازې د فعالیت د نمونې لپاره کارول شوی. د دوو ورکشاپونو لپاره، د حقیقي او deepfake نمونې منځنۍ نسبت 1: 1 دی. موږ د ورکشاپ لپاره کارول شوي ډاټا سیټونه یا د ډاټا سیټونو نور تفصيلي تخفیف نلري. د Deepfake ویډیو تشخیص ورکشاپ لپاره، د لومړي دوو ګټور ټیمونه (د A جایزه سره) د Netease 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 ). د دویمې کار لپاره "کروسګیال غږ بدلون" ګډون کوونکي باید د 24 VC سیسټمونو پراختیا وکړي، په شمول د نارینه او ښځو لخوا هم شامل دي، مګر په سه زبانونو کې (فینګ، آلمان او مینډینګ) په اساس د روزنې ډاټاټا جوړ کړي. انځور. د VC سیسټمونو روزنې او تولید پروسه ښيي. 90 91 92 93 4 6 Submissions were evaluated for “ ” لکه څنګه چې سازمانونه کارول [ ] 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 subjective evaluation 70 د Deepfake افریقا چیلنج (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 د DeepFake لوبې رقابت (DFGC) is in its first edition, hosted at the 2021 International Joint Conference on Biometrics (IJCB 2021). Its organisers are mainly from the Institute of Automation Chinese Academy of Sciences (CASIA). The idea of the competition was to promote an adversarial game between agents pushing for advances in both deepfake creation and detection. In order to achieve this, a 6-stage protocol was designed interleaving three creation phase (C-phase) and detection phase (D-phase), typically one week apart; submissions closed in April 2021. Both C-phases and D-phases were bound to the Celeb-DF (v2) dataset [ ]، چې 6229 ویډیوونه (590 حقیقي / اصل ویډیوونه او 5 639 جعلي / manipulated ویډیوونه) شامل دي، د روزنې لپاره. لکه څنګه چې د C فاز ته ورسیږي، به د Celeb-DF (v2) څخه د ډاټا سیټونو څخه جوړ شي، کوم چې د ارزونې پایلې ترلاسه کولو لپاره نوي چمتو کولو لارښوونې شامل دي. د D فاز ته ورسیږي به د تشخیص ماډلونو / کوډونو څخه جوړ شي چې د ارزونې پایلې ترلاسه کړي. د D فاز لپاره ورسیږي ماډلونه د مخکښ C فاز لپاره وړاندې شوي ډاټا سیټونو په پرتله ارزښت شوي دي [ ]. 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 is the number of valid submissions of created synthesis test sets in the last C-phase. DS 21 NC The Creation Score ( ) metric used to evaluate creation models submitted to this challenge is calculated by Eq. ( ), where 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 [ , , د ]. دا په سیستماتیک ډول د لاندې اړخونو پوښښ کوي: د تعریفونو او پراختیا، د کړنو معیارونه، ډاټا سیټونه، چټکونه / رقابتیونه / معیارونه، د کړنو پرتله، اصلي چټکونه او سپارښتنې. 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 , among researchers, practitioners and law makers there is no universally accepted definition of “deepfake” as a term. This is also reflected in how the authors of the 12 survey papers considered this aspect. Most authors talked about the history of deepfakes and pointed out that the term reflects the combination of “deep learning” and “fake”, but some used a broader definition, e.g., 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 [ , , , ] considered only videos, and only authors of [ د ] 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 [ , , ] even limited the definition of “deepfake” to such a narrow scope: 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. We call on more researchers to accept a broader definition of “deepfake” so that highly realistic/natural media of any kind generated by a sophisticated automated method (often AI-based) is considered deepfake. Here, we provide two examples of such a broader definition: the image2image (or pixel2pixel) technique [ ] 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 [ ]" چې د AI پر بنسټ تکنالوژیونه کارول کیږي چې واقعي ښکلي ټیټیټ انځورونه تولید کړي. 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 د فعالیت مټريکونه په حیرانتیا سره، د 12 سروې څخه هیڅکله د فعالیت د نمونې په مستقیم ډول پوښل نه شوې. ځینې په مستقیم ډول د فعالیت د نمونې لپاره کارول شوي دي ترڅو د پوښل شوي deepfake تولید او څیړنې لارښوونې د کارولو او پرتله کړي. د کارولو تر ټولو کارول شوي فعالیت د نمونې شامل دي دقت، ERR، او AUC. دا کیدای شي د دغو سروې کاغذونو صفحې محدودیتونو له امله وده ورکړي، کوم چې د انټرنېټانو ته اجازه نه ورکړي چې د فعالیت د نمونې په عمده توګه وده ورکړي. د deepfake د شخصي کیفیت د سروې لخوا لږ تر لږه پوښل شوي سيمه ده، کوم چې د فعالیت د ارزونې او مقایسه په اړه د deepfake تولید او deepfake څیړنې غیر متوازن پوښ 6.3 Datasets Many of the 12 survey papers list a number of deepfake-related datasets, but none of them have coverage as complete as ours shown in Section د بیلګې په توګه، د سروې څخه هیڅکله د 2016/2018/2020 Voice Conversion Challenge ډاټا سیټ نه پوښل شوی، او ASVspoof 2019/2021 ډاټا سیټونه یوازې په دوو سروې کې مختصر ډول پوښل شوي دي [ , ]. 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 [ ] covered many general fake image datasets where the manipulated images were not generated by deep learning or even AI-based methods, and some surveys (e.g., [ ]) 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 ) د سروې له خوا نه پوښل شوي دي. د چټکونو، رقابتیونو او بینکینګونو د تفصيلاتو کچه هم معمولا محدود دی، په پرتله چې موږ په برخه کې شامل کړي . 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 [ ]،، دوی هم د مختلفو ډولونو deepfake کشف روشونو په عمومي ځانګړتیاوو او ستونزوونو کې وګورئ، لکه څنګه چې په جدول کې ښيي . 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 د ډیرو deepfake-related ډاټا سیټونو د کیفیت ستونزو له امله (د برخه کې بحث شوي) ), 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. Not surprisingly, how to develop more robust, scalable, generalisable and explainable deepfake detection methods is one of the most discussed key challenges and also a major future research direction [ , , , د , , , د ]. د deepfake توليد او تشخیص تر منځ د سړک رخصتۍ په نظر کې، دا څیړنې لارښوونه به احتمالا د deepfake څیړنې کې ترټولو ګرم موضوع وي. 7 16 38 41 45 59 65 66 71 A couple of surveys [ د ] mentioned fusion as a key future research direction, where “fusion” refers to combining different methods (e.g., combining multiple detectors of different types) and data sources (e.g., jointly considering audio-visual analysis) to achieve better performance for deepfake detection. Lyu [ ] 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 د درې سروې جوړونکو، Lyu [ ] , Deshmukh and Wankhade [ ) او یوونوس او هاسان [ ]، argueed that better (higher-quality, more up-to-date, and more standard) deepfake datasets are needed to develop more effective deepfake detection methods. ] هم د دې وړاندیز کوي چې موږ اړتیا لرئ چې د په روزنې ډاټا اغیزو او د ډاټا سیټونو تبادلې ښه کړي. موږ په دې پوښونو کې د دوی سره موافق. 45 16 71 45 social media laundering 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. [ ] discussed specific research directions for different deep-fake generation methods (face synthesis, identity swap, attribute manipulation, and expression swap). Liang et al. [ ] and Li et al. [ ] 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. This paper provided a comprehensive overview of multiple aspects of the deepfake ecosystem drawing from the research literature and other online sources published in two languages: English and Chinese. It covers commonly used performance metrics and standards, related datasets, challenges, competitions and benchmarks. It also presents a meta-review of 12 selected deepfake-related survey papers published in 2020 and 2021, covering not only the above mentioned aspects, but also highlighting key challenges and recommendations. 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 and Steve Renals. 2018. Word Error Rate Estimation for Speech Recognition: e-WER. In . 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 . Springer, 17 pages. 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. 4، 9 (2020)، 283-292 د کمپيوټر علوم 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, and Lijun Yin. 2020. FakeCatcher: Detection of Synthetic Portrait Videos using Biological Signals. (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 صفحات. 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، او 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 او Sunil B. Wankhade. 2021. Deepfake د تشخیص لارښوونه د عمیق زده کړې په کارولو سره: د سیستماتیک جائزې. په . Springer, 293–302. Intelligent Computing and Networking: Proceedings of IC-ICN 2020 (Lecture Notes in Networks and Systems, 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، او مایکل Hahsler. 2020. د عمیق زده کړې او مسلکي ارزیابی په کارولو سره د چټک تشخیص بدلون. 2020, 1 (2020), 1–12. د معلوماتو د خوندیتوب په اړه د EURASIP مجله 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, and 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 n.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، او Jon Fiscus. 2021. د NIST رسنیو فورینسیک چیلنج (MFC) ډاټا سیټونو لپاره د کاروونکي لارښود. 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, and Ziwei Liu. 2021. ForgeryNet: A Versatile Benchmark for Comprehensive Forgery Analysis. In . IEEE, 4360–4369. د 2021 IEEE / CVF کنفرانس د کمپیوټر بصیرت او نمونې تشخیص په اړه پروسه 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 [33] Hyeongwoo کیم، Mohamed Elgharib، Hans-Peter Zoll ̈ofer، Michael Seidel، Thabo Beeler، Christian Richardt، او Christian Theobalt. 2019. د عصري سبک ساتنه د بصری ډبلنگ. 38, 6, Article 178 (2019), 13 pages. 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، قانون 163 (2018)، 14 مخونه. د Graphics په اړه د ACM سوداګرۍ https://doi.org/ 10.1145/3197517.3201283 [35] Pavel Korshunov and S´ebastien Marcel. 2019. Vulnerability Assessment and Detection of Deepfake Videos. In . IEEE, 1–6. Proceedings of the 2019 International Conference on Biometrics 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 48922.2021.01057 [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, and Siwei Lyu. 2018. In Ictu Oculi: Exposing AI Created Fake Videos by Detecting Eye Blinking. In . IEEE, 1–7. د 2018 IEEE نړیوال ورکشاپ د معلوماتو جاسوسی او امنیت په اړه پروسه 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, and Kai Chen. 2020. A Survey of Audiovisual Deepfake Detection Techniques. 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، او 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, and Zhenhua Ling. 2018. The Voice Conversion Challenge 2018: Promoting Development of Parallel and Nonparallel Methods. In . International Speech Communication Association, 195–202. د Odyssey 2018 پروژې 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. د 2020 IEEE نړیوال کنفرانس د Multimedia Expo ورکشاپونه https://doi.org/10.1109/ICMEW46912.2020.9105991 [46] Yisroel Mirsky او Wenke Lee. 2021. د Deepfakes جوړولو او کشف: د سروې. 54, 1, Article 7 (2021), 41 pages. د ACM کمپیوټر سروې 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), 1038–1048. IEEE Journal of Selected Topics in Signal Processing https://doi.org/10.1109/JSTSP.2020.3007250 [49] NIST Media Forensics Challenge ټیم. 2021. د Open Media Forensics Challenge 2020-2021 تحلیل پلان. 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، او Koray Kavukcuoglu. 2016 WaveNet: د Raw آډیو لپاره د جنراتور ماډل. 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, د 1391724 (2018)، 25 مخونه. International Journal of Digital Multimedia Broadcasting https://doi.org/10.1155/20 18/1391724 [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، او Wanyi Zhuang. 2021. DFGC 2021: د DeepFake لوبې رقابتی. arXiv:2106.01217. https: //arxiv.org/بس/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، او Weiming Zhang. 2020. DeepFaceLab: د انټرنېټ، انعطاف او پراختیا وړ چڼاسکه. https://doi.org/10.48550/ARXIV.2005.05535 [54] Stanislav Pidhorskyi, Donald A. Adjeroh, and Gianfranco Doretto. 2020. Adversarial Latent Au-toencoders. In IEEE، 10 صفحات. د 2020 IEEE / CVF کنفرانس د کمپیوټر بصیرت او نمونې تشخیص په اړه پروسه https://doi.org/10.1109/CVPR42600.2020.01411 [55] Andreas R ̈ossler، Davide Cozzolino، Luisa Verdoliva، Christian Riess، Justus Thies، او Matthias Nießner. 2018. FaceForensics: د انسانو په صورتونو کې د جعليې تشخیص لپاره د ویډیو ډاټاټایټ په پراخه کچه. https://doi.org/10.48550/ARXIV.1803.09179 [56] Andreas R ̈ossler، Davide Cozzolino، Luisa Verdoliva، Christian Riess، Justus Thies، او Matthias Nießner. 2019. FaceForensics++: زده کړې چې د مخابراتو په صورت کې انځورونه کشف کړي. په IEEE، 1 - 11 د کمپيوټر بصیرت 2019 نړیوال کنفرانس پروسه https://doi.org/10.110 9 / ICCV.2019 009 [57] Anindya سن. 2021. هنر او مصنوعي انټرنټ: څنګه د 'د Deepfakes' کولی شي د انټرنټ موزیم تجربو جوړولو کې مرسته وکړي! Medium بلاګ. 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، او Nobukatsu Hojo. 2019. WaveCycleGAN2: د وخت د ډومین د نوريال Post-فیلر لپاره د سپیک اول فارم نسل. https://doi.org/10.48550/A RXIV.1904.02892 [59] Jianhua Tao, Ruibo Fu, Jiangyan Yi, Chenglong Wang, and Tao Wang. 2020. Development and Challenge of Speech Forgery and Detection. 5، 2 (2020)، 28-38 Journal of Cyber Security http://jcs.iie.ac.cn/xxaqxb/ch/reader/view abstract.aspx?file no=20200204&flag=1 [60] Tencent. 2020 د مصنوعي انټرنټ سفید کتاب. https://tech.sina.com.cn/roll/2020-07-14/doc-iivhvpwx5201226.shtml [61] Justus Thies، Michael Zollh'ofe، او Matthias Niessner. 2019. د عصري رینډینګ: د عصري textures په کارولو سره د انځور synthesis. 38، د 66 (2019) ماده، 12 صفحات. ACM Transactions on Graphics https://doi.org/10.1145/3306346.3323035 [62] Tomoki Toda، Ling-Hui Chen، Daisuke Saito، Fernando Villavicencio، Mirjam Wester، Zhizheng Wu، او Junichi Yamagishi. د 2016 کال د غږ بدلون چیلنج. . International Speech Communication Association, 1632–1636. Proceedings of 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، او Kong Aik Lee. 2019. ASVspoof 2019: د راتلونکي هیرزونونه په مخکښ او جعلي آډیو تشخیص کې. arXiv:1904.05441. https://arxiv.org/ pdf/1904.05441.pdf [64] Ruben Tolosana، Ruben Vera-Rodriguez، Julian Fierrez، Aythami Morales، او Javier Ortega-Garcia. 2020. Deepfakes او په پرتله: د چڼاسکه manipulation او د جعلي څیړنې څیړنه. 64 (2020)، 131–148 د معلوماتو د Fusion https://doi.org/10.1016/j.inffus.2020.06.014 [65] Xin Tong، Luona Wang، Xiaoqin Pan، او Jingya Wang. 2020. د Deepfake په اړه د څارنې: په AI کې د Damocles شمشیر. . د IEEE، 265-273. د 2020 نړیوال کنفرانس په اړه د کمپیوټر بصیرت، انځور او عمیق زده کړې https://doi.org/10.1109/CVIDL51233.2020.00-88 [66] Luisa Verdoliva. 2020. د رسنیو فورینسیک او DeepFakes: د نندارتون. 14، 5 (2020)، 910-932 IEEE Journal of Selected Topics په سیگنال پروسس کې 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, Hirokazu Kameoka, Ingmar Steiner, Driss Matrouf, Jean-Fran¸cois Bonastre, Avashna Govender, Srikanth Ronanki, Jing-Xuan Zhang, and Zhen-Hua Ling. 2020. ASVspoof 2019: A Large-scale Public Database of Synthesized, Converted and Replayed Speech. 64 (2020), 27 pages. د کمپيوټر خبرې & ژبه https://doi.org/10.1016/j.csl.2020.101114 [68] Mirjam Wester، Zhizheng Wu، او Junichi Yamagishi. 2016. د Voice Conversion Challenge 2016 تبادله پايلې. په . International Speech Communication Association, 1637–1641. د Interspeech 2016 کنفرانس پروژې 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, and Andreas Nautsch. 2019. ASVspoof 2019: Automatic Speaker Verification Spoofing and 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، او Tomoki Toda. 2020. Voice Conversion Challenge 2020 - د زبانونو په منځ کې د نیمه پارامیل او د زبانونو په منځ کې Voice Conversion -. . International Speech Communication Association, 80–98. د Blizzard Challenge & Voice Conversion Challenge 2020 یوځای ورکشاپ پروژې https://doi.org/10.21437/VCC BC.2020-14 [71] Mohammad A. Younus او Taha M. Hasan. 2020. د Deepfake ویډیو څیړنې تکنالوژۍ د مختصر نظر. په IEEE، 115-120 د 2020 کال د 6th نړیوال انجنیري کنفرانس پروسه https://doi.org/10.1109/IEC49899.2020.9122916 [72] Guangtao Zhai and Xiongkuo Min. 2020. Perceptual Image Quality Assessment: A Survey. 63, Article 211301 (2020), 52 pages. Science China Information Sciences https://doi.org/10.1007/s11432-019-2757-1 [73] Teng Zhang، Lirui Deng، Liang Zhang، او Xianglei Dang. 2020. Deep Learning in Face Synthesis: د Deepfakes په اړه د سروې. په IEEE، 67 - 70 Proceedings of the 2020 IEEE 3rd International Conference on Computer and Communication Engineering Technology https://doi.org/10.1109/CCET5090 1.2020.9213159 په اړه [74] Yuxuan Zhang، Huan Ling، Jun Gao، Kangxue Yin، Jean-Francois Lafleche، Adela Barriuso، Antonio Torralba، او Sanja Fidler. 2021. DatasetGAN: سره د لږ تر لږه انسان هڅو اغیزمن لیبل شوي معلوماتو فابريکه. په IEEE، 10140-10150. د 2021 IEEE / CVF کنفرانس د کمپیوټر بصیرت او نمونې تشخیص په اړه پروسه 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، او Ziwei Liu. 2020. CelebA-Spoof: په پراخه کچه د فحش ضد ډاټا سیټ سره د غټ نوټونو. په سپینګر، 70-85 د 2020 کال د کمپیوټر بصری اروپايي کنفرانس پروسه 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، او Zhanlong Hao. 2021. CelebA-Spoof Challenge 2020 په چمتو کې Anti-Spoofing: روشونه او پایلې. arXiv:2102.12642. https://arxiv.org/pdf/2102.12642.pdf [77] Bo Zhao، Shaozeng Zhang، Chunxue Xu، Yifan Sun، او Chengbin Deng. 2021. Deep Fake جغرافیا؟ کله چې جغرافیایی معلوماتو د مصنوعي انټرنټ سره راځي. 4، 4 (2021)، 338-352 Cartography او جغرافیایی معلوماتو څیړنې https://doi.org/10.1080/15230406.2021.1910075 [78] Peng Zhou، Xintong Han، Vlad I. Morariu، او Larry S. Davis. 2017. د ټیمرټ چیرې د تشخیص لپاره دوه ټیم نوري شبکې. په IEEE، 1831-1839 د 2017 IEEE کنفرانس د کمپیوټر بصری او د نمونې تشخیص ورکشاپونه https://doi.org/10.1109/CVPRW.2017.229 [79] Tianfei Zhou، Wenguan Wang، Zhiyuan Liang، او Jianbing Shen. 2021. په وحشي کې فورینسیک. په IEEE، 5774-5784 د 2021 IEEE / CVF کنفرانس د کمپیوټر بصیرت او نمونې تشخیص په اړه پروسه https://doi.org/10.1109/CVPR46437.2021.00572 [80] Jun-Yan Zhu، Taesung پارک، Phillip Isola، او Alexei A. Efros. 2017. د بیلابیلو انځور-to-صور ترجمه په کارولو سره د دوامداره مخالفې شبکې. په IEEE، 2242-2251 د 2017 IEEE نړیوال کنفرانس د کمپیوټر بصیرت https://doi.org/10.1109/ICCV.2 017.244 [81] Bojia Zi، Minghao Chang، Jingjing Chen، Xingjun Ma، او Yu-Gang Jiang. 2020. WildDeepfake: د Deepfake تشخیص لپاره د واقعي نړۍ د ډاټا سیټ. په ACM، 2382–2390 د 2020 کال د 28th ACM نړیوال کنفرانس د Multimedia په اړه https://doi.org/10.1145/339417 1.3413769 دا کاغذ د CC by 4.0 Deed (Attribution 4.0 International) لائسنس لاندې archiv کې شتون لري. This paper is under CC by 4.0 Deed (Attribution 4.0 International) license. په Archive کې شتون لري