Seattle · multimodal ML · TH Cho

I teach machines to listen.
Then I make things worth hearing.

Multimodal ML engineer working on speech recognition, speaker identity, and egocentric perception for AI wearables — and a recording artist building a bilingual noir-cyberpunk record under the name TH Cho. Same obsession, two outputs: audio.

Selected work

Haven — offline-first disaster response with on-device Gemma

Kaggle Gemma 4 Good hackathon entry: an AI-d Kit that works when the network doesn't.

0 bytes network required

Per-event LoRA adapters for Whisper ASR

Domain-adapted speech recognition for AI smart glasses — entity-level WER cut by 15% with synthetic-only training data.

9.91% → 8.46% entity WER (raw)5.58% WER canonicalized

Listen

MAVERICK

mix & master · rap · production — catalog updates straight from Spotify

TH Cho — solo artist page lands July 7, 2026 with POP OUT!

Latest release