Speechmatics ↔ Twilio Voice — Reference Integration

A local proxy that bridges a Twilio Voice phone call to Speechmatics Realtime and prints per-speaker transcripts to your console. Both sides of a bridged two-party call are transcribed via Speechmatics' multi-channel diarization — you get a clean Inbound vs Outbound split with no client-side audio mixing.

This repo ships two identical-behaviour implementations. Pick the one that matches your stack and follow the README in that folder:

Language	Folder	Start here
Python	python/	python/README.md
Node.js / TypeScript	js/	js/README.md

Architecture

  ┌──────────┐      ┌────────┐      ┌──────────┐
  │ Caller A │◀──══▶│ Twilio │◀──══▶│ Caller B │      ═══  bridged voice
  └──────────┘      └────┬───┘      └──────────┘           (PSTN)
                         │
                         │  audio fork  (both_tracks, μ-law 8 kHz)
                         ▼
                 ┌───────────────┐
                 │ proxy.py /    │
                 │ proxy.ts      │
                 └───────┬───────┘
                         │
                         │  Realtime WebSocket
                         ▼
             ┌───────────────────────┐
             │ Speechmatics Realtime │
             │ (multi-channel)       │
             └───────────┬───────────┘
                         │
                         │  per-channel transcripts
                         ▼
                    ┌─────────┐
                    │ console │
                    └─────────┘

• Caller A dials the Twilio number; Twilio bridges the call to Caller B via TwiML <Dial> — two humans talk normally.
• Twilio simultaneously forks the live call audio (both tracks) to the proxy over WebSocket in Twilio's native μ-law 8 kHz format.
• The proxy relays those frames to Speechmatics Realtime — no transcoding, no re-encoding.
• Speechmatics returns per-channel transcripts tagged with the track they came from; the proxy prints them to the console with partials overwriting in place and finals committing to new lines.

The language-specific READMEs cover prerequisites, setup, .env variables, what to expect at runtime, tests, and troubleshooting.