Previously a Senior Research Scientist@Google Speech, and Senior Research Scientist@Google Research.Research on multi-task training for end-to-end speech recognizers. - Phoneme and grapheme joint modeling - Knowledge transfer from large language models to automatic speech recognizer (ASR) - Knowledge distillation and compact model estimationDeveloper of Pax framework (https://github.com/google/paxml) - mainly contributed to ASR-related components and modelsContributor of optax library (https://github.com/google-deepmind/optax) - added connectionist temporal classification (CTC) loss functionContributor of Jax library (https://github.com/google/jax) - added short-time Fourier transform (STFT) and inverse-STFT implementation(Side project) main developer of flacenc-rs (https://github.com/yotarok/flacenc-rs) - One of the fastest FLAC encoder implemented in Rust

- Development of modeling pipeline for voice search ASR products- Development of ASR models for mobile apps and set-top boxes- Internationalization of voice assistant ASR models- Research on compression of neural nets

Research on WFST-based structured classification methodResearch on application of Deep Neural Networks and Restricted Boltzmann Machines

Research on computationally efficient approximation of kernel-based classifiers/ probabilistic models

Research on Bayesian discriminative training method for generative statistical models

Research on multi-stream speech recognition systems