Show HN: SpeechCompass – speaker direction for captioning group conversations

Category: other

Tags: accessibility, speech-to-text, sound-localization, diarization, captcha, microphone-array, mobile-app, dsp

Score: 7.3/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 7)

SpeechCompass enhances mobile captioning by using multi-microphone localization and classical DSP to provide speaker direction and diarization in group conversations, addressing a key accessibility need for hard-of-hearing users. It combines custom hardware with portable C algorithms and an Android app, and includes validated user research and a best paper award at CHI 2025. The project is notable for its practical, low-compute approach that avoids voice data retention.

Target audience: backend devs, mobile devs, accessibility researchers, hardware engineers

Repository: https://github.com/google-deepmind/speech-compass/ · Java · Apache-2.0 · 3 stars

View on Hacker News