Microsoft Teams and Skype are used daily by hundreds of millions of users, and their usage has increased significantly since the COVID-19 pandemic and is a critical tool for working remotely and communicating with friends and family. In this talk we describe how we are replacing traditional digital signal processing components with machine learning based models. One recent new feature in Microsoft Teams and Skype for removing annoying background noise in telecommunication calls, which is the third most common call quality issue users complain about. We used deep learning to create a noise suppressor that performs >7X better than the previous non-machine learning solution. It’s a great feature, but how we developed it is even more interesting. Starting just under two years before shipping the feature, we first created three open source datasets and test sets for Deep Noise Suppression (DNS), as well as a best-in-class open source subjective test framework. We held two international challenges for DNS at INTERSPEECH 2020 and ICASSP 2021. Using the challenge results and our own models, we created the first background noise objective function that is highly correlated to human perception (PCC=0.97). This allowed us to iterate fast in model training and evaluation, and enabled us to create best in class DNS models. This type of open development model is new at Microsoft, and we are successfully applying it to another speech enhancement components like acoustic echo cancellation and packet loss concealment.

AI and deep learning has radically advanced many speech and audio processing applications. For example, we have all experienced improvements in speech recognition and synthesis in conversational systems on our phones. Recently deep learning has also revamped the quality of data compression of speech. In this talk we will discuss properties of some of these new architectures, and how they can be utilized to build a practical system for RTC.

In this presentation we discuss the various features and techniques that make libaom AV1 encoder suitable for RTC applications: from encoding tool selection to reducing complexity of the existing encoding tools. By implementing this approach libaom AV1 encoder is now able to provide low-latency real-time video encoding in RTC applications.

In this talk, we will mainly focus on the state-of-the-art AV1 software encoding capability for its deployment in RTC use cases, taking our Aurora1 AV1 as an instantiation. RTC in essence requests a sufficiently fast encoder with limited computational resource requested, so that it can be deployed in a wide range of client devices with minimum delay incurred. Meanwhile, the encoder should also collaborate closely with the network layer to handle varying and various networking conditions. AV1 as a codec standard provides several intrinsic critical features, making it a better fit for the RTC scenario. For instance, for the first time AV1 includes the Screen Content Coding (SCC) tools, such as IntraBC, palette, etc., in its main standard body, ideal for the coding of screen content that is widely present in such RTC use cases as video conferencing, distance learning, and remote desktop sharing. Taking our Aurora1 AV1 encoder as an example, we will first demonstrate AV1 is quite ready and has already been deployed for RTC. For SCC, for instance, Aurora1 is able to conduct a steady 1080p real-time encoding in zero latency at speed of 15fps, using one single CPU core on a PC platform, consuming just a quarter or even less of the bitrate consumed by x264 superfast whereas providing a superior visual quality. Aurora1 can also encode 720p30 camera content real time and steadily over recent generations of mobile platforms, e.g. iPhone12, with a significant coding efficiency advantage as opposed to existing codec standards. We will present such advantages of AV1 for RTC with extensive, detailed stats data, comparing Aurora1 against widely acknowledged open source encoders, including OpenH264, x264, libvpx-VP8/VP9, SVT-AV1, and liabom RT (libaom real-time mode). We will also introduce AV1's scalability support, as well as its natural integration with WebRTC across various PC and mobile platforms, optimized for both x86/ARM architectures. We will illustrate different speed presets provided by Aurora1 and confirm that more aggressively faster speed levels are being developed. We will also describe the role that is being played by dav1d, an open source AV1 decoder led by the joint efforts of VideoLAN and FFmpeg, funded by AOM, together with the ecosystem of AV1 that is being gradually formed.

The users of Meta RTC products experience a very diverse set of network conditions, some of those may be far from perfect. In this presentation, we are going to cover the following discussions: (1) the challenges we faced to deliver smooth real-time audio over lossy network conditions; (2) the error resiliency technologies we explored to recover audio packet losses; (3) the delicate trade-offs between audio packet loss recovery and end-to-end latency; (4) future directions to improve user perceptive audio quality with packet loss.

WhatsApp’s mission is to connect the world privately by designing a product that's simple and private. Privacy and security is in our DNA. In this presentation, we are going to talk about how we build the private calling for billions of users by leveraging client centric architecture and E2EE. We'll start from discussing the protection levels of user privacy, and our basic E2EE and architecture for 1:1 call and the challenges and benefits for privacy in a client centric E2EE design. Then we will talk about how we evolve our E2EE and architecture for group call and multi-device for scalability.

Meta helps billions of users connect daily by providing real time communication services. Group call is one of these services were more than two users can join a call. A common way of supporting group calls is by leveraging Selective Forwarding Unit (SFU) server to route the media among the participants. In an e2e encrypted call, the server forwards the media without having the capability to decrypt them. In this talk we will discuss the methods to encrypt the media communicated among the participants and the encryption key negotiation mechanism. We will also discuss how to keep the call secrecy under various scenarios such as participants joining or leaving the call. As the call size grows (more participants in a call), there are increasing number of the challenges to keep the e2e encryption and the call secrecy under different scenarios. We will discuss these challenges and ways to solve them.