PeopleofSuperintelligence

Researcher

Daniel Povey

Also known as Dan Povey

Daniel Povey is a speech recognition researcher whose work includes Kaldi and neural acoustic-model training. After earning a Cambridge PhD in 2003, he worked at IBM Research and Microsoft Research, then spent seven years at Johns Hopkins. His work includes LF-MMI sequence training and Zipformer, a transformer encoder for automatic speech recognition. ISCA named him a Fellow in 2026; as of October 2026, he is Xiaomi Group's Chief Voice Scientist.

Timeline

  1. 2003–2012

    From Cambridge to industry

    He completed a PhD at Cambridge in 2003 with a thesis on discriminative training for large-vocabulary speech recognition. He spent about ten years in industrial research at IBM Research and Microsoft Research, working at Microsoft until early 2012.

  2. 2009–2011

    The Kaldi project begins

    Kaldi grew from a 2009 Johns Hopkins workshop and its code was released in 2011. After its release, he helped maintain and develop much of the toolkit with other contributors.

  3. 2012–2019

    Training speech models at Hopkins

    At Johns Hopkins, he served as non-tenure-track faculty from 2012 to 2019. In 2016, he and coauthors described LF-MMI sequence training for neural acoustic models without frame-level cross-entropy pretraining.

  4. 2019–2023

    A lab in Beijing

    In November 2019, he moved to Beijing to join Xiaomi as Chief Voice Scientist. The Zipformer encoder, which he coauthored in 2023, is a faster, more memory-efficient transformer for automatic speech recognition.

  5. 2026

    A fellowship for speech work

    ISCA selected him as a 2026 Fellow for contributions to speech recognition algorithms, data and software.

  6. As of October 2026

    Leading Xiaomi's speech work

    He is Xiaomi Group's Chief Voice Scientist.

Sources and credits

Photo: We have no published photo of Daniel Povey yet. Send us a free-licensed photo.

  1. The Challenge of Good Documentation for Machine Learning Software cfcs.pku.edu.cn. Checked 14 October 2026.
  2. Kaldi: History of the Kaldi project kaldi-asr.org. Checked 14 October 2026.
  3. Dan Povey danielpovey.com. Checked 14 October 2026.
  4. Purely Sequence-Trained Neural Networks for ASR Based on Lattice-Free MMI isca-archive.org. Checked 14 October 2026.
  5. Zipformer: A faster and better encoder for automatic speech recognition arxiv.org. Checked 14 October 2026.
  6. 小米语音首席科学家 Daniel Povey 当选 2026 ISCA Fellow，全球仅 100 余人获此殊荣 m.ithome.com. Checked 14 October 2026.
  7. 顶尖人才 | 小米招聘 hr.xiaomi.com. Checked 14 October 2026.
  8. NewX | Research Hub newx.sg. Checked 14 October 2026.