[Example] 410 — Sinch Voice API Call Transcription with Deepgram STT

[github-actions]

New example: Sinch Voice API Call Transcription with Deepgram STT

Integration: Sinch Voice API | Language: Node.js | Products: STT

What this shows

A Node.js Express server that handles inbound Sinch phone calls via the Voice API, responds with SVAML to route call audio over a WebSocket (ConnectStream), and forwards raw linear16 PCM audio to Deepgram for real-time transcription. Demonstrates the full pipeline: Sinch ICE/ACE/DiCE webhooks → ConnectStream WebSocket → Deepgram live STT → console transcript output.

Required secrets

• SINCH_APPLICATION_KEY — from Sinch Voice dashboard
• SINCH_APPLICATION_SECRET — from Sinch Voice dashboard

Tests

✅ Tests passed

Downloading test audio...
Converting to linear16 16 kHz mono...
Audio ready: 829866 bytes of linear16 16 kHz

Server started on :3098
POST /sinch/ice → valid SVAML with connectStream
  destination: ws://localhost:3098/stream
  sampleRate: 16000
POST /sinch/ace → valid response with continue action
GET / → {status: "ok"}

Streaming audio through server → Deepgram (up to 30 s)...
[stream] Sinch WebSocket connected
[sinch] Stream metadata — call: ci-test-call
[deepgram] Connection opened
[interim] Yeah. Is
[final] Yeah. As as much as, it's worth
[final] celebrating the first spacewalk, with an all female team,

Received 9 transcript event(s)
Transcript length verified: 323 chars for 10.0s audio

All tests passed

Closes #142

---

Built by Engineer on 2026-04-03

[github-actions]

⏳ Waiting on credentials

Missing: SINCH_APPLICATION_KEY SINCH_APPLICATION_SECRET

Add these as repository secrets and the next review sweep will pick this up.