PeopleofSuperintelligence

Researcher

Szu-Wei Fu

Also known as 傅思維

Szu-Wei Fu is a speech researcher whose MetricGAN method trains speech-enhancement models against evaluation metrics. MetricGAN appeared at ICML in 2019, and he finished a computer science PhD at National Taiwan University in 2020 under Shou-De Lin and Yu Tsao. From 2020 he worked on the SpeechBrain project at Mila with Yoshua Bengio and Mirco Ravanelli, and in 2021 he moved to Microsoft as a data and applied scientist. He joined NVIDIA Research in November 2022 and is a Senior Research Scientist there, working on speech and multimodal AI as a member of NVIDIA Research Taiwan.

Now at
NVIDIA
Works
88
Citations
3,660
h-index
23

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2019

    MetricGAN at ICML

    In 2019, he is first author of MetricGAN, accepted at ICML, with Chien-Feng Liao, Yu Tsao and Shou-De Lin. The approach optimizes a generator against one or several evaluation metrics, and the paper tests it on speech enhancement.

  2. 2020

    PhD at National Taiwan University

    He completes a PhD in computer science at National Taiwan University in 2020, advised by Shou-De Lin and Yu Tsao. His thesis, "Investigation of Cost Function for Task-Oriented Speech Enhancement", studies loss functions for training speech-enhancement models.

  3. 2020–2021

    SpeechBrain at Mila

    In May 2020, he joins the SpeechBrain project at Mila in Canada, advised by Yoshua Bengio and Mirco Ravanelli. In 2021, he is one of the 21 authors of the paper that describes SpeechBrain, an open-source all-in-one speech toolkit.

  4. 2021

    MetricGAN+ for speech enhancement

    At Interspeech 2021, he presents MetricGAN+ with Mirco Ravanelli, Yu Tsao and colleagues. Three added training techniques raise the PESQ score by 0.3 over MetricGAN, to 3.15 on the VoiceBank-DEMAND dataset.

  5. 2021–2022

    Applied scientist at Microsoft

    In June 2021, he joins Microsoft as a data and applied scientist, and he works there as a senior applied scientist in Vancouver until his move to NVIDIA Research in 2022.

  6. November 2022

    Joining NVIDIA Research

    He joins NVIDIA Research in November 2022. His interests there include ML-based audio-visual processing, speech processing and enhancement, and quality estimation.

  7. 2025

    VoiceNoNG and DeSTA2.5-Audio

    At Interspeech 2025, he coauthors VoiceNoNG, a speech-editing model that applies latent flow matching to the pre-quantization features of a neural codec. In July 2025, he is also among the 28 authors of DeSTA2.5-Audio, a large audio language model whose backbone LLM generates its own training targets.

  8. As of October 2026

    Senior Research Scientist at NVIDIA

    He is a Senior Research Scientist at NVIDIA Research, working on speech and multimodal AI. He is a member of NVIDIA Research Taiwan, founded in 2022 as the APAC research team of NVIDIA Research, with a focus on generative AI, multimodal learning and 3D vision.

Career

  1. Since 2022NVIDIASenior Research Scientist
  2. 2021–2022MicrosoftData and Applied Scientist; Senior Applied Scientist
  3. 2020–2021Mila
  4. 2016–2022Academia Sinica

Education

  1. 2020National Taiwan UniversityPhD

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

  1. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Szu-Wei Fu. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.
  2. Google has the highest combined h-index among AI organizationsIncludes NVIDIA. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  3. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes NVIDIA. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.
  4. Universities (31%) and startups and other companies (25%) employ the largest shares of 2020–2026 entrantsIncludes NVIDIA. Among 819 people whose first recorded role started in 2020–2026 and who have a known current organization, 60% work in commercial sectors (25% at startups and other companies, 23% at frontier labs and 13% at big tech, with 9 in two), and 31% work at universities, compared with 66% for earlier entrants.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Szu-Wei Fu yet. Send us a free-licensed photo.

  1. MetricGAN: Generative Adversarial Networks based Black-box Metric Scores Optimization for Speech Enhancement arxiv.org. Checked 20 October 2026.
  2. Szu-Wei Fu's Web jasonswfu.github.io. Checked 20 October 2026.
  3. NTU Theses and Dissertations Repository: Investigation of Cost Function for Task-Oriented Speech Enhancement tdr.lib.ntu.edu.tw. Checked 20 October 2026.
  4. SpeechBrain: A General-Purpose Speech Toolkit arxiv.org. Checked 20 October 2026.
  5. MetricGAN+: An Improved Version of MetricGAN for Speech Enhancement isca-archive.org. Checked 20 October 2026.
  6. Szu-Wei Fu | NVIDIA Research Taiwan research.nvidia.com. Checked 20 October 2026.
  7. Szu-Wei Fu | Research research.nvidia.com. Checked 20 October 2026.
  8. VoiceNoNG: Robust High-Quality Speech Editing Model without Hallucinations isca-archive.org. Checked 20 October 2026.
  9. DeSTA2.5-Audio: Toward General-Purpose Large Audio Language Model with Self-Generated Cross-Modal Alignment arxiv.org. Checked 20 October 2026.
  10. NVIDIA Research Taiwan research.nvidia.com. Checked 20 October 2026.