Alexander Richard

Alexander Richard is a research scientist at Meta Reality Labs in Pittsburgh whose work on audio-visual learning supports lifelike avatars and social telepresence in virtual reality. He studied at RWTH Aachen University, wrote a 2019 PhD thesis on temporal segmentation of human actions in videos at the University of Bonn, and joined Facebook Reality Labs in October 2019. In 2021, he published MeshTalk, a method that animates a full 3D face from speech, and Neural Synthesis of Binaural Speech From Mono Audio, which received an ICLR Outstanding Paper Award. By 2022, he led the audio-visual research team at Reality Labs Research, and his later work includes From Audio to Photoreal Embodiment in 2024 and the Seamless Interaction dataset in 2025.

Now at: Meta
Badges: Bridge
Links: Website, Google Scholar and X

Timeline

2008–2014

Degrees at RWTH Aachen

He studies at RWTH Aachen University, earning a bachelor's degree in 2011 and a master's degree in 2014. His master's thesis, Improved Optimization of Neural Networks, adds a neural network module to the RASR speech recognition software and evaluates a new optimizer, mean-normalized stochastic gradient descent (MN-SGD).
2014–2019

PhD on action segmentation

He is a PhD student at the University of Bonn from 2014 to 2019, working on temporal action segmentation and action labeling in long, untrimmed videos. In 2017, he presents an oral at CVPR on weakly supervised action learning with RNN-based fine-to-coarse modeling, written with Hilde Kuehne and Juergen Gall. His 2019 PhD thesis is titled Temporal Segmentation of Human Actions in Videos.
2018–2019

Joining Reality Labs in Pittsburgh

In 2018, he does a research internship at Facebook Reality Labs in Pittsburgh on multi-modal modeling for social VR. After a stint as a scientist on Amazon Alexa in Aachen, Germany, in 2019, he joins Facebook Reality Labs, now Meta Reality Labs, in Pittsburgh as a research scientist in October 2019.
2021

MeshTalk and an ICLR award

He is first author of MeshTalk, a 2021 method that generates full facial 3D animation from speech. His paper Neural Synthesis of Binaural Speech From Mono Audio, which turns a single-channel audio source into two-channel binaural sound in real time, receives an ICLR 2021 Outstanding Paper Award. In August 2021, he discusses his Codec Avatar work at Facebook Reality Labs on the TWIML AI Podcast.
2022

Leading the audio-visual team

In September 2022, he gives a VASC Seminar at Carnegie Mellon's Robotics Institute, Audio-Visual Learning for Social Telepresence, as the research scientist leading the audio-visual research team at Reality Labs Research in Pittsburgh. His team works on audio-visual learning for photo- and phono-realistic immersive experiences in virtual reality.
2024

From audio to photoreal avatars

In 2024, he coauthors From Audio to Photoreal Embodiment: Synthesizing Humans in Conversations with Evonne Ng, Javier Romero, Timur Bagautdinov, Shaojie Bai, Trevor Darrell and Angjoo Kanazawa. From the audio of a two-person conversation, the system generates photorealistic face, body and hand gestures by combining vector quantization with diffusion, and it appears at CVPR 2024.
2025

Seamless Interaction dataset

He is one of 84 authors of Seamless Interaction: Dyadic Audiovisual Motion Modeling and Large-Scale Dataset, posted in June 2025, which presents over 4,000 hours of face-to-face interaction footage from over 4,000 participants and models that generate gestures and facial expressions aligned with speech. He also coauthors AV-Flow in 2025, an audio-visual generative model that animates photo-realistic 4D talking avatars from text input alone.
As of October 2026

Research scientist at Reality Labs

He is an AI research scientist at Meta Reality Labs in Pittsburgh, working on multimodal modeling for social VR. A SIGGRAPH Asia 2026 paper, Audio Driven Real-Time Facial Animation for Social Telepresence, lists him among its authors and animates photorealistic 3D facial avatars from audio in real time.