Deep Learning Researcher - Berkeley Speech Group
Current
- Constructed new interpretable representations for universal foundational speech models based on multimodal articulatory measurement data spanning across audio, EMA, rt-MRI, X-ray, EMG, and more - Independently developed a novel point recognition algorithm for low-resource training using Kullback–Leibler divergence of spatial probability maps for use in extracting multispeaker articulatory MRI representations, used in speech synthesis, avatar generation, and speech pathology; trained large-scale models with distributed parallel processing- Paper: Deep Articulatory MRI Feature Extraction - IEEE ICASSP 2024 (first author, under review), Multimodal Pretraining for Vocal Tract Modeling - CVPR 2024 (first author, under review)