Kureis in the Enterprise Collaboration Industry

Kure

On Some Crucial Points in Psychoanalysis

THE THEORY OF SEXUAL INVERSION

Studies in the Psychology of Sex, Volume 2 by Havelock Ellis is part of the HackerNoon Books Series. You can jump to any chapter in this book here . PREFACE TO FIRST EDITION PREFACE TO FIRST EDITION. It was not my intention to publish a study of an abnormal manifestation of the sexual instinct before discussing its normal manifestations. It has happened, however, that this part of my work is ready first, and, since I thus gain a longer period to develop the central part of my subject, I do not regret the change of plan. I had not at first proposed to devote a whole volume to sexual inversion. It may even be that I was inclined to slur it over as an unpleasant subject, and one that it was not wise to enlarge on. But I found in time that several persons for whom I felt respect and admiration were the congenital subjects of this abnormality. At the same time I realized that in England, more than in any other country, the law and public opinion combine to place a heavy penal burden and a severe social stigma on the manifestations of an instinct which to those persons who possess it frequently appears natural and normal. It was clear, therefore, that the matter was in special need of elucidation and discussion. There can be no doubt that a peculiar amount of ignorance exists regarding the subject of sexual inversion. I know medical men of many years' general experience who have never, to their knowledge, come across a single case. We may remember, indeed, that some fifteen years ago the total number of cases recorded in scientific literature scarcely equaled those of British race which I have obtained, and that before my first cases were published not a single British case, unconnected with the asylum or the prison, had ever been recorded. Probably not a very large number of people are even aware that the turning in of the sexual instinct toward persons of the same sex can ever be regarded as inborn, so far as any sexual instinct is inborn. And very few, indeed, would not be surprised if it were possible to publish a list of the names of sexually inverted men and women who at the present time are honorably known in church, state, society, art, or letters. It could not be positively affirmed of all such persons that they were born inverted, but in most the inverted tendency seems to be instinctive, and appears at a somewhat early age. In any case, however, it must be realized that in this volume we are not dealing with subjects belonging to the lunatic asylum, or the prison. We are concerned with individuals who live in freedom, some of them suffering intensely from their abnormal organization, but otherwise ordinary members of society. In a few cases we are concerned with individuals whose moral or artistic ideals have widely influenced their fellows, who know nothing of the peculiar organization which has largely molded those ideals. I am indebted to several friends for notes, observations, and correspondence on this subject, more especially to one, referred to as "Z.," and to another as "Q.," who have obtained a considerable number of reliable histories for me, and have also supplied many valuable notes; to "Josiah Flynt" (whose articles on tramps in Atlantic Monthly and Harper's Magazine have attracted wide attention) for an appendix on homosexuality among tramps; to Drs. Kiernan, Lydston, and Talbot for assistance at various points noted in the text; and to Dr. K., an American woman physician, who kindly assisted me in obtaining cases, and has also supplied an appendix. Other obligations are mentioned in the text. All those portions of the book which are of medical or medico-legal interest, including most of the cases, have appeared during the last three years in the Alienist and Neurologist, the Journal of Mental Science, the Centralblatt für Nervenheilkunde, the Medico-legal Journal, and the Archivo delle Psicopatie Sessuale. The cases, as they appear in the present volume, have been slightly condensed, but nothing of genuine psychological interest has been omitted. Owing to some delay in the publication of the English edition of the work, a German translation by my friend, Dr. Hans Kurella, editor of the Centralblatt für Nervenheilkunde, has already appeared (1896) in the Bibliothek für Sozialwissenschaft. The German edition contains some matter which has finally been rejected from the English edition as of minor importance; on the other hand, much has been added to the English edition, and the whole carefully revised. I have only to add that if it may seem that I have unduly ignored the cases and arguments brought forward by other writers, it is by no means because I wish to depreciate the valuable work done by my predecessors in this field. It is solely because I have not desired to popularize the results previously reached, but simply to bring forward my own results. If I had not been able to present new facts in what is perhaps a new light, I should not feel justified in approaching the subject of sexual inversion at all. HAVELOCK ELLIS. About HackerNoon Book Series: We bring you the most important technical, scientific, and insightful public domain books. This book is part of the public domain. Havelock Ellis (2004). Studies in the Psychology of Sex, Volume 2. Urbana, Illinois: Project Gutenberg. Retrieved October 2022 https://www.gutenberg.org/cache/epub/13611/pg13611-images.html This eBook is for the use of anyone anywhere at no cost and with almost no restrictions whatsoever. You may copy it, give it away or re-use it under the terms of the Project Gutenberg License included with this eBook or online at www.gutenberg.org , located at https://www.gutenberg.org/policy/license.html.

PREFACE TO FIRST EDITION

FORGETTING OF IMPRESSIONS AND RESOLUTIONS

Is it possible to achieve less than a second latency in a video broadcast? What if the stream goes to a thousand people? Yes. How? Let’s answer using the project as an example. We did it using WebRTC. WCL streams HD concerts to an audience of hundreds and thousands of people. WorldCast Live Why would I reduce the broadcast latency? Latency less than a second in a videoconference is normal, otherwise speaking is nearly impossible. For one-sided streaming, the latency of 2 or even 20 seconds is fine. For example, TV delay is about 6 sec. However, there are cases where you’d try to reduce it to 1 second. Sport events It’s highly unlikely that the user will be happy when their neighbors shout, GOAL!, and he still sees the ball somewhere in the middle of the field. What if the user is also live betting? Interviews Thanks to the pandemic, not only talks with friends have moved online, but also interviews with celebrities. Take an interviewer’s latency, add to the interviewee’s latency, and add the time spent while it all goes to the user. The more you get, the worse the experience is for all sides. Concerts The WCL player is embedded in different sites. Concerts broadcast online to all of them. If you and your friend use different sites to watch the same concert, it will negatively affect your viewing experience. Thus, the concert organizers are trying to minimize the latency. Although the user wouldn’t really care whether he hears the guitar riff now or in 2 seconds, there is a general trend on minimizing the latency. The faster the better! And also… The examples above combine. In WCL viewers and performers talk via a video chat. Therefore, the latency has to be like that of a video chat, as we need the minimum difference in time between the question and the answer. How to reach low latency in live streaming? Use WebRTC The standard WebRTC package offers average latency of 500 ms (half a second). We could’ve finished the article here: creating a video chat where people can connect with each other isn’t difficult. What’s difficult is making it all work stable for thousands of people and customizing streams to improve their quality. Base WebRTC isn’t about HD streaming. Video and audio quality is enough for speaking but isn’t enough for streaming music. To decrease the latency, WebRTC can reduce the quality and skip parts of the content. To avoid it, you have to go under the hood of WebRTC, which we did. Set up WebRTC To ensure that the low latency doesn’t go against quality and the final user’s experience, you need to go for extra development. Here’s what we did for WCL so it can broadcasts concerts to thousands of people: Enabled multi-channel audio The standard audio in WebRTC is mono, it’s 1 channel. The stereo is 2 channels. There are 5 channels on WCL. Upped bitrate WebRTC settings limit the video transmission speed to 500 kb/s. That’s not much for action on the screen: if light colors change quickly with this limit, the quality will be lower because of trying to go inside that channel, pixels might appear. Not a great watching experience. Therefore we’ve increased the bitrate to 1,5 Gb to transmit HD video. Increased discretization frequency By this, we’ve improved the quality of audio and video, low and high frequencies. They are not lost anymore. We can’t disclose what exactly we did, but if you’re interested in that, let us know! Scale Kurento is an open-source WebRTC server.  Set up Master Kurento to stream video. 500 hundred people will connect directly to Master Kurento, from where they’ll get the stream. If there are more viewers, Edge Kurento is in play, to which other viewers connect. The more viewers there are on the stream, the more Edge Kurento you need to use. All together it creates a tree-like scheme. Kurento Media Server When do you leave the latency of more than a second? When the budget is limited. if you need scaling. The WebRTC server on WCL costs $0,17/h which is $122,4 monthly if we take 30 days. WebRTC is more expensive than HLS HLS, however, costs $0,023/h and can be turned on and off, unlike WebRTC. If we take 3 hour-long concerts a week, then we’ll spend a bit less than $0,28 for 12 concerts. Note that there are many server providers, the prices are always different, but you can see what it looks like using our project as an example. The first stable version of subsecond latency on WCL took us 3,5 weeks. If there’s no necessity in very low latency, why spend? TL;DR Latency less than a second in a broadcast is necessary when there’s communication between participants or a stream where things change all the time. WebRTC makes it possible for even a thousand people if you work with the technology. If your app is about something from here, make sure to take a look at WebRTC. Contact us, too, we’ll help! Contact us via our or DM on , which we created not so long ago :) form Instagram Previously published at https://forasoft.com/blog/article/minimizing-latency-to-less-than-1-sec-for-mass-streams-315 This article was written by Nikolay Tarakanovkiy, Fora Soft Senior project manager

How to Minimize Latency to Less than 1 Second for Mass Streams

Speech-to-text (STT), also known as automated-speech-recognition (ASR), has a long history and has made amazing progress over the past decade. Currently, it is often believed that only large corporations like Google,  Facebook, or Baidu (or local state-backed monopolies for the Russian language) can provide deployable “in-the-wild” solutions. Original TOC Related Work and Inspiration Open Speech To Text (Russian) Making a Great Speech To Text Model Model Benchmarks and Generalization Gap Further Work Abstract Speech-to-text (STT), also known as automated-speech-recognition (ASR), has a long history and has made amazing progress over the past decade. Currently, it is often believed that only large corporations like Google,  Facebook, or Baidu (or local state-backed monopolies for the Russian language) can provide deployable “in-the-wild” solutions. This is due to several reasons: High compute requirements that are usually used in papers erect artificially high entry barriers; Speech requiring significant data due to the diverse vocabulary, speakers, and compression artifacts; A mentality where practical solutions are abandoned in favor of impractical, yet state of the art (SOTA) solutions. In this piece we describe our effort to alleviate these concerns, both globally and for the Russian language, by: Introducing the diverse 20,000 hour published under CC-NC-BY license; Open STT dataset Demonstrating that it is possible to achieve competitive results using only TWO consumer-grade and widely available GPUs; Offering a plethora of design patterns that democratize entry to the speech domain for a wide range of researchers and practitioners. Introduction Following the success and the democratization (the so-called , i.e. the reduction of hardware requirements, time-to-market and minimal dataset sizes to produce deployable products) of computer vision, it is logical to hope that other branches of Machine Learning (ML) will follow suit. The only questions are, when will it happen and what are the necessary conditions for it to happen? “ ” ImageNet moment In our opinion, the ImageNet moment in a given ML sub-field arrives when: The architectures and model building blocks required to solve 95% of standard “useful” tasks are widely available as standard and tested open-source framework modules; Most popular models are available with pre-trained weights; Knowledge transfer from standard tasks using pre-trained models to different everyday tasks is solved; The compute required to train models for everyday tasks is minimal (e.g. 1–10 GPU days in STT) compared to the compute requirements previously reported in papers (100–1000 GPU days in STT); The compute for pre-training large models is available to small independent companies and research groups; If the above conditions are satisfied, one can develop new useful applications with reasonable costs. Also democratization occurs — one no longer has to rely on giant companies such as Google as the only source of truth in the industry. If you would like to know more about the philosophy of our work and why we opted for online publications instead of conferences / peer reviewed papers - please follow this . link Related Work and Inspiration For our experiments we have chosen the following stack of technologies: Feed-forward neural networks for acoustic modelling (mostly grouped 1D convolutions with and transformer blocks); squeeze and excitation loss (CTC loss); Connectionist temporal classification Composite tokens consisting of graphemes (i.e. alphabet letters) as modelling units (opposed to phonemes); Beam search with a pre-trained language model (LM) as a decoder. There are many ways to approach STT. Discussing their drawbacks and advantages is out of scope here. Everything in this article is said about an end-to-end approach using mostly graphemes (i.e. alphabet letters) and neural networks. In a nutshell — to train an end-to-end grapheme model you just need a lot of small audio files with corresponding transcriptions, i.e. file.wav and transcription.txt. You can also use CTC loss, which alleviates the requirement to have time-aligned annotation (otherwise you will need either to provide an alignment table by yourself or learn alignment within your network). A common alternative to CTC loss is the standard categorical cross-entropy loss with attention, but it trains slowly by itself and it is usually used together with CTC loss anyway. This “stack” was chosen for a number of reasons: Scalability. You can scale your compute by adding GPUs; Future proofing. Should a new neural network block become mainstream, it can be integrated and tested within days. Migrating to another framework is also easy; Simplicity. Namely using Python and PyTorch you can focus on experimentation and not solving legacy constraints; Flexibility. Building proper code in Python you can test new features (i.e. speaker diarization) in days; By not using attention in the decoder nor phonemes or recurrent neural networks we achieve faster convergence and need less maintenance for our models; Open Speech To Text (Russian) All publicly available supervised English datasets that we know of are smaller than 1,000 hours and have very limited variability. , a seminal STT paper, suggests that you need at least 10,000 hours of annotation to build a proper STT system. 1,000 hours is also a good start, but given the generalization gap (discussed below) you need around 10,000 hours of data in different domains. DeepSpeech 2 Typical academic datasets have the following drawbacks: Too ideal. Recorded in studio or too clean compared to real world applications; Too narrow of a domain. Difficulty in STT follows this simple formula: noise level * vocabulary size * number of speakers; Mostly only English. Though projects like alleviate this constraint to some extent, you cannot reliably find a lot of data in languages other than German and English. Also Common Voice is probably more suitable for speaker identification task more than speech-to-text because their texts are not very diverse; Common Voice Different compression. Wav files have little to no compression artifacts and therefore don’t represent real world sound bytes that are compressed in different ways; Because of these drawbacks, about 6 months ago we decided to collect and share an unprecedented spoken corpus in Russian. We targeted 10,000 hours at first. To our knowledge this is unprecedented even for the English language. We have seen an to do work similar to ours, but despite the government funding, their datasets are not publicly available. attempt Recently we released a of the dataset. It includes the following domains: 1.0-beta version Our data-collection process was the following: Collect some data then clean it using heuristics; Train some models and use those models to further clean the data; Collect more data and use alignment to align transcripts with audio; Train better models and use those models to further clean the data;Collect more data and manually annotate some data; Repeat all the steps. You can find and you can . our corpus here support our dataset here Though this is already substantial, we are not yet done. Our short term plan is: Do some housekeeping, clean the data more, and clean-up some legacy code; Migrate to .ogg in order to minimize data storage space while maintaining quality; Add several new domains (courtroom dialogues, medical lectures and seminars, poetry). PS. We did all of this, our dataset was even featured on azure datasets, now we are planning in releasing pre-trained models for 3 new languages: English / German / Spanish. Making a Great Speech To Text Model To build a great STT model, it needs the following characteristics: Quick inference; Parameter-efficient;Easy to maintain and improve; Does not require a lot of compute to train, a 2 x 1080Ti machine or less should suffice; We take these as our goals, and describe how we fulfilled them below. Traditionally models are selected by benchmarking them on a couple of fixed “ideal” unseen validation datasets. In the previous sections we explained why this is sub-optimal if you have real world usage in mind and the only datasets available are academic datasets. Given limited resources to properly compare models you need a radically different approach, which we present in this section. Also keep in mind that there is no “ideal” validation dataset when you are dealing with real in-the-wild data — you need to . validate on each domain separately Usually when reporting some results on some public dataset (e.g. ImageNet), researchers allegedly run full experiments with different hyper-parameters from scratch until convergence. Also, a good practice is to run the so-called ablation tests, i.e. experiments that test whether or not additional features of a model were actually useful by comparing the performance of the model with and without those features. In real life, practitioners cannot afford themselves the luxury of running hundreds or thousands of experiments from scratch till convergence or building some fancy reinforcement learning code to control experiments. Also, the dominance of over-parameterized methods in the literature and the availability of enterprise oriented toolkits discourages researchers from deeply optimizing their pipelines. When you explore the hardware options, in the professional or cloud segment there is a bias towards expensive and impractical solutions. Read more to learn about our model selection methodology. here Overall Progress Made Initially we started with a fork of . The original Deep Speech 2 model is based on a deep LSTM or GRU recurrent network, which are slow. The above image illustrates the optimizations we were able to add to the original pipeline. More specifically, we were able to do the following without hurting model performance: Deep Speech 2 in PyTorch Reduce the model size around 5x; Speed up its convergence 5–10x; The small (25M-35M params) final model can be trained on 2x1080 Ti GPUs instead of 4; The large model still requires 4x1080 Ti but has a bit lower final CER (1–1.5 percentage point lower) compared to the small model. The above chart only has convolutional models, which we found to be much faster than their recurrent counterparts. We started on the process to getting these results as follows: Used an existing implementation of Deep Speech 2; Run a few experiments on LibriSpeech, where we noticed that RNN models are typically very slow compared to their convolutional counterparts; Added a plain Wav2Letter inspired model, which was actually underparameterized for Russian, so we increased the model size; Noticed that the model was okay, but very slow to train, so we tried to optimize the training time. So, we then explored the following ideas to improve things: Idea 1 — Model Stride; Idea 2 — Compact Regularized Networks; Idea 3 — Using Byte-Pair Encoding; Idea 4 — Better Encoder; Idea 5 — Balance Capacity — Never Use 4 GPUs Again; Idea 6 — Stabilize the Training in Different Domains, Balance Generalization; Idea 7 — Make A Very Fast Decoder; Please follow this to learn about each of these ideas in detail. link Model Benchmarks and Generalization Gap In real life it is expected that if the model is trained on one domain, there will be a significant generalization gap on another. But is there a generalization gap in the first place? If there is, then what are the main differences between domains? Can you train one model to work fine on many reasonable domains with decent signal-to-noise ratio? There is a generalization gap, and you can even deduce which ASR systems were trained on which domains. Also, with the ideas above, you can train a model that will perform decently even on unseen domains. According to our observations, these are the main differences that cause the generalization gap between domains: Overall noise level; Vocabulary and pronunciation; The codecs or hardware used to compress audio; This benchmark includes both an acoustic model and a language model. The acoustic model is run on GPU, the results are accumulated, and then language model post-processing is run on multiple CPUs For more detailed benchmarks, some thoughts on production usage and benchmark analysis please go . For up-to-date and updated benchmarks please go (Russian) here here . Further Work Here is a list of ideas, that we tested (some of which even worked), but we decided in the end that their complexity does not justify the benefits they provide: Getting rid of gradient clipping. Gradient clipping takes from 25% to 40% of batch time. We tried various hacks to get rid of it, but could not do it without suffering a severe drop in convergence speed; ADAM, Novograd and other new and promising optimizers. In our experience, they worked only with simpler non speech related domains or toy datasets; Sequence-to-sequence decoder, double supervision. These ideas work. Attention-based decoders with categorical cross-entropy loss instead of CTC are notoriously slow starters (you add speech decoding to the already burdensome task of alignment). Hybrid networks did not perform much better to justify their complexity. This probably just means that hybrid networks require a lot of parameter fine-tuning; Phoneme-based and phoneme-augmented methods. Though these helped us regularize a few over-parametrized models (100–150M params), they proved not very useful for smaller models. Surprisingly an extensive tokenization study by Google ; arrived at the similar result Networks that increase in width gradually. A common design pattern in computer vision, so far such networks converged worse that their counterparts with the same network width; Usage of . At first glance, this did not work, but maybe more time was needed to make it work; IdleBlocks Try any sort of tunable filters instead of STFT. We tried various implementations of tunable STFT filters and SincNet filters, but in most cases we could not even stabilize the training of the models with such filters; Train a pyramid-shaped model with different strides. We failed to achieve any improvement here;Use model distillation and quantization to speed up inference. At the moment when we tried native quantization in PyTorch it was still in beta and ; did not support our modules yet Add complementary objectives like speaker diarization or noise cancelling. Noise cancelling works, but it proved to be more of an aesthetic use; Author Bio Alexander Veysov is a Data Scientist in Silero, a small company building NLP / Speech / CV enabled products, and author of Open STT. Silero has recently shipped its own Russian STT engine. Previously he worked in a then Moscow-based VC firm and Ponominalu.ru, a ticketing startup acquired by MTS (major Russian TelCo). He received his BA and MA in Economics in Moscow State University for International Relations (MGIMO). You can follow his channel in telegram (@snakers41). Originally published at https://thegradient.pub on March 28, 2020. Acknowledgments Thanks to Andrey Kurenkov and Jacob Anderson from The Gradient for their contributions to this piece. Citation For attribution in academic contexts or books, please cite this work as Alexander Veysov, “Toward’s an ImageNet Moment for Speech-to-Text”, The Gradient, 2020. BibTeX citation @article{veysov2020towardimagenetstt, author = {Veysov, Alexander}, title = {Toward’s an ImageNet Moment for Speech-to-Text}, journal = {The Gradient}, year = {2020}, howpublished = {\url{ https://thegradient.pub/towards-an-imagenet-moment-for-speech-to-text/ } }, } Originally published at https://thegradient.pub on March 28, 2020. All citations and references preserved as they were in the original article. HN also does not have handy table-of-contents features, so I will leave the original links as well. Where appropriate, I will provide a link to the original part of the article. I will also provide links to more up-to-date benchmarks. Also soon we will be releasing pre-trained models for English / German / Spanish.

Towards an ImageNet Moment for Speech-to-Text: A Deep Dive

A job with 50 flights a year, and a life in back to back audio and video call meetings, and the same question over and over again in my mind: Why are all these conferencing platforms so shit, we’ve got AI making deep fakes from dead people in 4K with Dolby surround, and my colleague is still an unintelligible single delayed pixel falling off the edge of the screen out of sync, and no one seems to ever be able to share the right screen. No I’m still seeing the Presenter View. I quit a dream job with a multinational technology company last year to try and fix this. This is not my first rodeo but I quickly found out the problem was a lot deeper and fundamental then I expected. Then Covid happened and now everyone’s asking the same question I was asking: Why are all these conferencing platforms so shit. Let me take you down the rabbit hole I find myself that’s all about choosing where you attack on the value stacks, market dynamics of an oddly sized fragmented market, and what I think could be one of the greatest business opportunities I’ve seen in ages. Analysis Paralysis There’s been numerous surveys done on the problems with video conferencing, you’ll know the list: , the list goes on, long and familiar. Audio issues, sync issues between video and voice, poor picture quality, latency, echo problems and hearing yourself, people not hearing you and not being able to interrupt them And then there’s been many physiological studies on why we have such a negative experience, from the effect of lack of eye contact suggesting dishonesty, to the overloading of the brain the division of people into little boxes on the screen causes, and the decremental effects of hearing yourself. So the Google’s and the Microsoft’s of this world, or the younger focused companies like Zoom, why haven’t the titans of technology, fixed these already? The Heart of the Problem The reason these tools are not great is actually really obvious if you’ve used a specialised commercial software or service, like your own homemade HR or holiday booking system in your company, or one of those websites for a small local government service for example, they’re all terrible to use, full of bugs, and don’t work well or as expected. The reason is there’s just not that big a user-base to justify the invest to build something better. Building robust reliable services with well thought out UX takes a lot of investment. Before Covid hit, the whole of the video conferencing market in the US was only worth about $3.85B (USD). It was split about 50/50 between HW and SW/Service. Also, corporate conferencing, the bit you first think of, is only about a quarter of the market again, with other submarkets like Education, Healthcare, Government and Defence making the other pieces. If you think the corporate platforms are not great, you should see the other specialised ones, they’re shockingly bad. So we’ve started with not that great a number, and divided it by 2, and then divided by 4, you’re now in the Which isn’t that big a number if you think about it. 100s of millions of dollars range. Then of course the market is fragmented, and each company has a smaller piece of this. When you get to those levels of revenue, quite frankly the return on investment on building say a new advance echo cancellation system isn’t great. If you looks at the teams behind many of these products, a large part of the operational cost is around the sales organisation, winning and keeping those customers, and the engineering teams are grossly under resourced and overloaded. Which brings me to the other characteristic of this market: Imagine you were using one of those virtual background capabilities that are very fashionable at the moment, that cut people out of their background and make them look like they’re on the beach (for some reason) and it didn’t recognise the face of your colleague that has darker skin properly, and they disappeared. You can imagine that happening, and that’s the exact reason you wouldn’t use that feature in a serious meeting. The need for extremely high reliability. For a capability to reach that enterprise level quality it needs to be truly well qualified for all edge scenarios, and all biases to be removed, and again that adds significant cost, and reduces the ROI further. And that’s the main reason these tools aren’t great. Making them better isn’t outside the roam of human capability, it’s just that they’re not worth it for the companies to do, all the best engineers in the world are off pushing the envelope on deep fakes, they don’t have time for this stuff. How to Fix it To find the solution we need to go back to those budget local government websites, and free service apps again. In the last 5 years some of them have got a lot better, Why? In big part this is due to the availability of various intermediate technology stacks that you can build your service on top of. Everyone knows you’re not gonna go out there now and buy your own servers, you’ll use AWS/GCP/Azure for the compute and serving capability, but a lot of the underlaying SW you need to deal with capabilities like accounts, billing, data management, security, and frameworks for UIs are either available for free, or there’s great paid for services, and they are so much better than anything you could ever make yourself, because they have the economy of scale. You’ll left to make the application SW only so your investment goes a lot further. The world of video and audio conferencing doesn’t have this, you can’t buy in all those well-developed underlying capabilities you need, well not yet anyway. You’d have to build a lot of it yourself. If you set out to build your own video conferencing platform tomorrow, say with a flow that was more usable, or say you wanted to focus and build something with certain features suitable for , you’d likely get the HW on AWS, you’ll use open source for the OS and server infrastructure, you’ll even be able to find a great white-label video conferencing SW platform, free or paid to take and tailor, but if you use these pieces and build your application on top of it, better screen sharing remote exercise classes you’d likely have poor audio quality, and echo problems, and disappointing video performance. If you wanted better quality there, even on par with the Zooms and Microsofts of this world, you’ll have to sink millions into making your own audio/video signal processing, since there isn’t anything out there free or paid for that’s any good and properly qualified, and yes big guys like Microsoft, Google and Zoom have their own, but they’re not gonna sell just that capability to you to build a competing platform to them, and there’s isn’t even that good, because they couldn’t justify the investment needed in the first place (as explained above). Go a bit more application specific, that background swap feature, you’d have to build your own and it to make sure it works for all kinds of webcams, and rooms shapes, and backgrounds, and skin tones, etc. That’s a reasonable investment, and that’s only to get you on par with the incumbents. fully qualify So unfortunately despite your best intentions your platform will be just as bad, if not worse, and even though you have better screen sharing flow, or suitable market features, people will not want to use your product. The way to break this situation is that we need independent technology companies that want to invest and do the audio and video signal processing well, and their business model is setup to sell these capabilities to the conferencing platform providers, new and old. At this point, the solution should be obvious: The benefit of this approach is that . If you develop a great real time noise cancelling capability, AI powered, highly qualified, you can sell this to all the major conferencing providers, and not just corporate, but also education, healthcare, and even other markets outside of conferencing. Now the market size is big enough to justify the investment, and you’re not then trying to build, market and sell a whole conferencing platform to end users, but just this capability. the ROIs start making sense If you look at the conferencing market, and want to make it better, It won’t be better, you just won’t have pockets deep enough, and even if you did, you’ll never get a reasonable return on it. Try not to be the hero and household name, instead identify the underlying technologies that would make the existing platforms better or enable better new platforms, develop them, and find a business model that allows you to sell them to many. I believe the worst thing you could do is to try and make another conferencing platform. , there would be major independent technology companies at various levels of value stack. There would be 2 or 3 companies offering great noise suppression capabilities, some would also have echo cancelling, there would a number of technology companies offering great real-time video processing capabilities, and there would be startup’s breaking through with new advance capabilities for this market, some quite high up the value stack with very application specific features. If I were to describe a world where video conferencing isn’t rubbish As an end user of conferencing you wouldn’t be buying these capabilities from these companies, in fact , but the conferencing platform providers you use and know, would buy these capabilities from the companies mentioned above, and this would free the platform companies to use those sales organisations to listen to the end customers, figure out what they need at the application level, and tailor products to work the way the customers want, with access to rich powerful underlying technologies from various suppliers. you would likely never hear the name of these companies at all To fix this market we need an ecosystem of players, that is the solution to the problem, not one vertical behemoth owning the whole stack. We’ve seen that doesn’t work. The Change is Coming With the explosion in video conferencing due to Covid-19, the market size has obviously grown, but it’s too early to know how much of this change and size increase is here to stay, but beside the market size growing and allowing the big players to justify their investments on technology, finally, independently we’re started to see more companies trying to build the underlaying technologies with a sell to all business model, rather than making their own platform. On noise suppression a great example is a startup called , building an AI power noise suppressor. Other recent examples are a tool in beta called from All Turtles, which is a set of features mostly around video and image processing to make the experience of video calls more productive. Krisp mmhmm Today these companies are offering their capabilities as plugins that the end user can use with any conferencing platform. I think it feels inevitable that for these startups to succeed they will have to evolve out of offering plugins to end users and become technology suppliers to the platform providers, but that in itself is a journey they will have to go through. This is more than just Video Conferencing You might be thinking some of this real time audio signal processing capabilities must have existed for years, my phone does it, surely not every phone manufacturer is going trying to build their own capabilities from scratch? If you go back 5 years, the majority of the audio signal processing was happening on the communication devices themselves, like your phone, or your physical conference phone sitting in the conference room, and even on your laptop, and traditional integrated circuit and signal processing companies like Analog Devices, Texas Instruments, STMicroelectornics, have been providing real time signal processing capabilities for decades. But two things have happened recently, that changes things up: - While the initial signal processing in the devices remain, we’re seeing a emergence of the more advance signal processing capabilities away from the devices themselves to the cloud, where there’s more information, or information from multiple devices, enabling much better processing opportunities for certain problems - Advances in AI for both audio and video processing, combined with the vast increase in centralised processing capability has increased the possibilities but has also pushed the signal processing further up from the embedded devices in many cases and away from the offerings of the traditional incumbents in the real-time signal processing space. I believe and hopeful that we’ll witness : Either the traditional power houses of signal processing, predominately the IC manufactures, will grow their AI driven capability and transition further up the signal processing chain, and offer more application specific audio and video processing capability, or There’s a real need for this. one of 2 things over the next decade in this space of real-time signal processing we’ll see a new breed of major company emerge, filling that gap in the ecosystem, providing application specific real-time signal processing capabilities. So What Am I Doing About All This? I have a new startup, but we’re still in stealth, so we’re not talking much about what we’re doing at the moment, but we have a couple of cool bits technology and our goal is to build state-of-the-art, fully qualified audio and video signal processing capabilities, with a focus in the voice and video communication and conferencing market, no surprises there. We want to make video conferencing better, a lot better. We’re experimenting with a sort of business model, where you pay for the real time processing of audio and video feeds in the cloud in a by minute model, this way regardless of your size and application you can get hold of the best signal processing needed for your application, and you also don’t need to worry about the infrastructure or latency, it’s all taken care of. Digital Signal Processing as a Service (DSPaaS) We’re hoping to flip this market on its head, and if we do our job right, you’ll never hear our name, but your video conference platform, whichever one you use, and whatever you use it for, should suddenly get a whole lot better. References If you want to learn and read more on the space, here’s some links to the studies, market analysis and companies I talked about in the article above: A nice survey by OWLLabs on the the state of video conferencing in 2018, and some commentary on what some of the biggest challenges in using video conferencing is OWLLabs — State of Video Conferencing 2018 https://www.owllabs.com/state-of-video-conferencing Some commentary in the article on study of the effect of lack of eye contact in video conferencing, and how it causes lack of trust Ksenia Klykova — The impact of videoconferencing on business travel: an historical point of view https://medium.com/sam/the-impact-of-videoconferencing-on-business-travel-an-historical-point-of-view-3b5fa85e0a9d New feature from Microsoft Teams, but interestingly their study in seeing that having people in different boxes on the screen seems to make you more tired Tom Warren — Microsoft Teams’ new Together Mode is designed for pandemic-era meetings https://www.theverge.com/2020/7/8/21317526/microsoft-teams-together-mode-dynamic-view-new-features If you want to find out more information about the video conferencing market sizes and segmentations Grand View Research — Video Conferencing market Size, Share 7 Trends Analysis Report https://www.grandviewresearch.com/industry-analysis/video-conferencing-market Another look at the video conferencing market, with sub market breakdown — Corporate, Education, Healthcare, etc Fortune Business Insights — Video Conferencing Market Size, Share & Industry Analysis, By Type, By Application, By Enterprise Size and Regional Forecast https://www.fortunebusinessinsights.com/industry-reports/video-conferencing-market-100293 There’s a few, but OpenVidu is an free or paid for white label video conferencing platform, based on Kurento, that you can take and build your own video conferencing platform on top of pretty quickly for example OpenVidu and Kurento https://openvidu.io https://www.kurento.org Krisp, is that noise suppression startup I was talking about above. They have a free plugin you can try Krisp https://krisp.ai mmhmm is that beta tool that help with screen sharing and processes the video on calls, on their website at the moment at least you can find a video explaining what they’re doing. I’ve added the link for the video as well. mmhmm https://www.mmhmm.app https://www.youtube.com/watch?v=c8KhKBLoSMk This is our new startup company working in this space, there’s almost no information on the website at the moment, sorry. Pommelhorse https://pommelhorse.com

Why is Video Conferencing so SH*T - Surely we can do better than this

Plaza’s MerchantChain — Next Gen Commerce on Blockchain What is happening now with Crypto and blockchain technology is how I imagine the internet gold rush happened in the 90’s felt. Since I was too young to experience those years, I am paying extra close attention to what is happening today. The concept behind blockchain technology is trust and transparency. Blockchain will have a massive impact on digital identity, banking, smart contracts, decentralized notary, digital voting and many more. In current scenario sellers need middleman like Amazon, flip-kart, eBay and Alibaba for selling and promote their products all over the world. For selling their products approximately 25% fees paying for e-commerce vendors, that cost they will add in their product cost and finally consumers are paying more than the actual price. Nowadays traditional payments systems are using all over the world for e-commerce transactions and for that also transactions fees we are paying extra. Cryptocurrency payments can be used for such type of transactions, however for the average person who has never purchased anything with cryptocurrency, if he makes a simple mistake can result in total loss. For e-commerce the idea of a fully decentralized marketplace powered by blockchain is a great goal. Plaza systems will eliminate such costs help sellers as well as enable them to sell directly to the end consumer and cutting down the unnecessary middlemen. Plaza’s Systems provides worlds fastest (around 15,000 +tps) , most sustainable, most scale-able and most secure decentralized e-commerce infrastructure that gives every consumer & Sellers can get benefit from blockchain technology. Plaza-systems having advantages from both the centralized and decentralized software solutions also. Plaza Systems is generating their revenues from marketplace fees as an e-commerce meta search engine and their major revenue coming from referral fees. Plaza Systems is currently developing the Total b Commerce solution, which includes the fast and future-proof Merchant-chain — a blockchain commerce infrastructure upon which others can build decentralized applications — as well as the Freedom Lifestyle. Plaza systems is developing both a limited-supply utility token — Plaza and a stable, tethered transactions coin (PL$) for buying goods and services in the Plaza marketplace. The PlazaCard™ helping to crypto owners to use their cryptocurrency for everyday payments in shops and restaurants all over the world. Merchant-chain are entire business intelligence layer executed on dApps, not servers; fully autonomous. Plaza systems are made some successful partnership like; Leading online crypto news portal The Currency Analytics, Digital wallet, e-commerce and tokenised reward platform developer KartBlock, Community-managed gambling platform U Run has signed a memorandum of understanding with Plaza systems. AgentMile, the world’s first decentralized commercial real estate leasing platform powered by AI, has signed a “mutual investment” deal with Plaza systems. Light-emitting diode (LED) proponent Terawatt and Plaza systems Plaza are “joining forces” and “pooling resources” in an agreement that leverages the strengths of both projects. Largest cryptocurrency project from Africa KureCoin has chosen to build out its banking, trading, and payments solution on Plaza’s Merchantchain and issue a fixed-value crypto based on the MerchantChain Transaction Coin Standard. Plaza Systems already raised from ICO 43,350 ETH and ICO ends December 27, 2018. Token info BONUS PLAZA Presale Bonuses: Week 1 March 15–22 — > 50% bonus (reserved tokens only)Week 2 March 22–29 — > 30% bonus (reservations expired)Week 3 March 29-August 15 — > 25% bonus PLAZA Mainsale Bonuses: Tier 1 Day 1 Sept 15 — > 15% bonus (1 ETH = 5,750 PLAZA)Tier 2 Sep 16–Sept 23 — > 10% bonus (1 ETH = 5,500 PLAZA)Tier 3 Sept 23–Sept 30 — > 5% bonus (1 ETH = 5,250 PLAZA)Tier 4 Sept 30-Oct27 — > no bonus (1 ETH = 5,000 PLAZA) Road map: Plaza System ICO Rating: “This article was created in exchange for a potential token reward through Bounty0x” Username — Narasimman1987 For more information: Plaza | Freedom Lifestyle DocSend Document Plaza Discussion Group PLAZA (@PlazaSystems) | Twitter

KURE

kureko.com

-- employees

Since n.d.

COMPANY RANKING

KURE

EVERGREEN INDEX #5809

Representing this company? Click here to claim and customize this page!

Kure WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!

KURE

kureko.com

-- employees

Since n.d.

COMPANY RANKING

KURE

EVERGREEN INDEX #5809

RELATED COMPANIESRANK

STORIES ABOUT KURE

Representing this company? Click here to claim and customize this page!

Kure WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!