ANCHOR — speech quality from partial audio
Predicts how listeners will rate speech before the utterance finishes, with 21–35% lower PLCMOS error than ARECHO on 2-second prefixes.
I'm Terry. I work with Prof. Shinji Watanabe at WavLab on recovering speech from damaged audio and predicting how it sounds to listeners. I'm currently the lead student researcher on an Amazon AGI project on target-speaker speech restoration.
Seeking Summer 2027 research or ML engineering internships in speech and audio
Predicts how listeners will rate speech before the utterance finishes, with 21–35% lower PLCMOS error than ARECHO on 2-second prefixes.
@inproceedings{tao26_interspeech,
title = {{ANCHOR: Autoregressive Non-intrusive Chunk-Ordered Refinement for Joint Multi-Resolution Speech Quality Modeling}},
author = {Zhuoyan Tao and Jiatong Shi and Hye-jin Shim and Shinji Watanabe},
year = {2026},
booktitle = {{Interspeech 2026}},
pages = {5672--5677},
doi = {10.21437/Interspeech.2026-927},
issn = {2958-1796},
}
A benchmark for when conversational systems should speak or wait, with 30 hours of hand-labeled conversation and 14 systems benchmarked.
@inproceedings{jiang2026turnbench,
title = {{TurnBench}: A Multi-Domain Benchmark for Turn-Taking Dynamics in Spoken Dialogue},
author = {Freeman Jiang and Ramon Sanabria and Soham Deshmukh and Bandhav Veluri and Simon Michael Vuch Williams and Elliott K. Suen and Garreth Lee and Kevin Yoonho Choi and Takuya Umeki and Riku Kubo and Sathvik Udupa and Chien-yu Huang and Shih-Yun Shan Kuan and Zhuoyan Tao and Satyapriya Krishna and Sefik Emre Eskimez and Yu Tsao and Hung-yi Lee and Shinji Watanabe},
booktitle = {Proc. IEEE Spoken Language Technology Workshop (SLT)},
year = {2026},
eprint = {2608.25218},
archivePrefix = {arXiv},
}
A single training-free number, computed from language-ID posteriors, that forecasts whether fine-tuning a diarization probe will pay off.
@misc{tao2026offtarget,
title = {Off-target posterior mass: A training-free diagnostic for fine-tuning spoken language diarization probes},
author = {Zhuoyan Tao and Aditya Kommineni and Shrikanth Narayanan},
year = {2026},
eprint = {2611.13430},
archivePrefix = {arXiv},
primaryClass = {eess.AS},
}
M.S. in Intelligent Information Systems, expected Dec 2027. Speech research at WavLab with Prof. Shinji Watanabe.
Built simulation tooling with ~1000× higher experiment throughput.
Trustee Scholar (full-tuition merit scholarship); graduated magna cum laude. Multilingual speech research with Prof. Shrikanth Narayanan at SAIL.
Outside research, I led the audio team for Bone Boy, a game released on Steam, and taught multivariable calculus review sessions. Full experience and honors in my résumé ↗︎
For research collaborations or Summer 2027 internships: terrytao@andrew.cmu.edu.