PeopleofSuperintelligence

Researcher

Alexander H. Liu

Also known as Alexander Liu

Alexander H. Liu is an AI researcher whose work spans computer vision, speech representation learning, and generative audio. He earned B.S. and M.S. degrees in computer science and information engineering at National Taiwan University and a Ph.D. at MIT under James Glass. His research includes SeqRQ-AE, DinoSR, and SpeechFlow, plus Mistral AI's Voxtral audio models and Magistral reasoning model. As of October 2026, he is a Member of Technical Staff at Thinking Machines Lab.

Works
39
Citations
825
h-index
13

Publication numbers from OpenAlex, as of 15 October 2026.

Timeline

  1. 2018–2019

    Early computer vision

    His 2018 NeurIPS paper proposes a unified feature disentangler for multi-domain image translation and manipulation. At CVPR in 2019, he coauthors SceneNet, a method for unsupervised monocular depth estimation that uses semantic understanding.

  2. 2020

    Speech from unpaired data

    In 2020, he co-first-authors SeqRQ-AE, which learns phoneme-like units from unpaired audio for speech recognition and synthesis. The method uses under 20 minutes of labeled speech in its recognition and synthesis experiments.

  3. 2022

    Unsupervised speech synthesis

    In 2022, he co-first-authors Simple and Effective Unsupervised Speech Synthesis, which uses unlabeled audio, text, and a lexicon to produce intelligible speech without a labeled corpus.

  4. 2023

    DinoSR speech learning

    In 2023, his NeurIPS paper DinoSR combines masked language modeling, self-distillation, and online clustering to learn speech representations. The model forms a phone inventory from audio embeddings and uses those units to train a student network.

  5. 2024

    SpeechFlow pretraining

    In 2024, he coauthors SpeechFlow, a flow-matching model pretrained on 60,000 hours of untranscribed speech. Fine-tuning adapts it to speech enhancement, separation, and synthesis.

  6. 2025

    Mistral audio and reasoning models

    At Mistral AI, he coauthors Voxtral, multimodal audio chat models for spoken audio and text. He also coauthors Magistral, Mistral AI's first reasoning model.

  7. By 2026

    Voxtral systems and MIT training

    By 2026, he holds B.S. and M.S. degrees in computer science and information engineering from National Taiwan University, and a Ph.D. in computer science from MIT under James Glass. In February 2026, he coauthors Voxtral Realtime, a streaming speech-recognition model built for sub-second latency. In March 2026, he coauthors Voxtral TTS, a multilingual model that generates natural speech from as little as three seconds of reference audio.

  8. As of October 2026

    Member of Technical Staff

    As of October 2026, he is a Member of Technical Staff at Thinking Machines Lab. His research focuses on natural language and speech processing for AI.

Sources and credits

Photo: We have no published photo of Alexander H. Liu yet. Send us a free-licensed photo.

  1. Alexander H. Liu alexander-h-liu.github.io. Checked 15 October 2026.
  2. Towards Scene Understanding: Unsupervised Monocular Depth Estimation With Semantic-Aware Representation openaccess.thecvf.com. Checked 15 October 2026.
  3. [1910.12729] Towards Unsupervised Speech Recognition and Synthesis with Quantized Speech Representation Learning arxiv.org. Checked 15 October 2026.
  4. [2204.02524] Simple and Effective Unsupervised Speech Synthesis arxiv.org. Checked 15 October 2026.
  5. DinoSR: Self-Distillation and Online Clustering for Self-supervised Speech Representation Learning papers.neurips.cc. Checked 15 October 2026.
  6. [2310.16338] Generative Pre-training for Speech with Flow Matching arxiv.org. Checked 15 October 2026.
  7. [2506.10910] Magistral arxiv.org. Checked 15 October 2026.
  8. [2507.13264] Voxtral arxiv.org. Checked 15 October 2026.
  9. [2602.11298] Voxtral Realtime arxiv.org. Checked 15 October 2026.
  10. [2603.25551] Voxtral TTS arxiv.org. Checked 15 October 2026.