Integrate with Twilio

Bring low‑latency, high‑quality speech from Async into any Twilio voice call. This guide shows how to:

Connect to Async Text‑to‑Speech (WebSocket)

Spin up a local WebSocket server for Twilio <Stream/> media

Expose that server through ngrok

Dial a phone call and pipe the generated audio into it

1 Prerequisites

Tool	Notes
Node.js 18+	ES module syntax and modern WS APIs
Twilio account	Copy your Account SID and Auth Token; buy/verify numbers
Async account	Copy your API key and pick a Voice ID
ngrok (free)	Exposes your local WS server to Twilio’s cloud

Create a .env file next to the script:

You can also supply any of these via command‑line, e.g. node async-twilio.js OUTBOUND_NUMBER=+1555….

Step	Flow
1	Script connects to Async over WebSocket, sends an init frame (model, voice, codec).
2	A lightweight HTTP + WS server starts locally (`ws://localhost:<port>`).
3	`ngrok` publishes that port; you get a public wss:// URL.
4	Script tells Twilio to dial `<OUTBOUND_NUMBER>` and stream call audio to that URL.
5	On Twilio `start`, script streams text → Async.
6	Async replies with μ‑law PCM chunks; script forwards each chunk to Twilio as `media` frames.
7	After all chunks (or on timeout) script ends the call.

Setup the required variables and helper functions

Connect to async TTS

Local ws setup for twilio

Pipe async → twilio

Twilio helper functions

Running the functionality

Goal	Where to change
Different voice	`CFG.ASYNC_VOICE_ID`
Different codec / rate	`output_format` in `connectAsyncTTS()`
Stream arbitrary text	Replace `CFG.TEST_SENTENCE`, or feed user input into `asyncWs.send()`
Keep the call open	Remove the `chunksSeen` guard and `endCall()` timer

No audio? Make sure Twilio can reach your ngrok URL (port 443, wss).

Choppy playback? Forward Async chunks to Twilio as soon as they arrive—don’t buffer them.

Delay before speech starts? Use force: true in the transcript frame to synthesize short text immediately.

Bidirectional audio – Send caller speech to Async STT and build IVR bots.

Fail‑over logic – Retry with a new ngrok tunnel or switch data centres automatically.

Serverless deployment – Move the bridge to AWS Lambda or Fly.io and drop ngrok for a fixed WSS endpoint.