jxlarrea/wyoming-voice-match
A Wyoming protocol ASR proxy that verifies speaker identity and isolates voice commands from background noise before forwarding audio to a downstream speech-to-text service. Designed for Home Assistant voice pipelines to prevent false activations from TVs, radios, and other people - and to deliver clean transcripts even in noisy environments.
Implements speaker verification and extraction using ECAPA-TDNN neural networks to identify enrolled voices and isolate them from background audio in real-time. Operates as a Wyoming protocol proxy that buffers post-wake-word audio, performs voiceprint matching within ~5 seconds, then applies energy-based speech segmentation to filter and reconstruct clean audio containing only the target speaker before forwarding to downstream ASR services.
Stars
34
Forks
1
Language
Python
License
MIT
Category
Last pushed
Mar 11, 2026
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/voice-ai/jxlarrea/wyoming-voice-match"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.