Enrollment-free target speech extraction
Pulling out the person you're engaging with — without asking them for a reference clip first — so the system stays out of the way of the conversation.
12.83 dB SI-SDRi · ongoing
PhD · Soundability Lab · University of Michigan
Sidharth
Teaching machines to hear the one voice you're talking to.
Speech & audio AI + HCI. I build target speech extraction, speaker representations and real-time enhancement models that work in the noisy rooms where conversations actually happen. Advised by Dhruv Jain, with Hao-Wen Dong.
↳ move over the room: the voice nearest your cursor is extracted
SUMMER 2026Spatial audio ML intern
at Bose Research
at Bose Research
Soundability Lab
