MyLab TTS Streaming API

Synthesize speech in a registered voice and receive the audio as it is generated, over a single WebSocket connection.

Overview

A WebSocket connection is a session, intended to span an entire call: open it when the call starts, send one synthesize request per agent turn, and close it when the call ends. Inactivity does not close a session.

Prerequisites: an API key issued to your account (a secret; store it only on your server) and a voice ID (voiceProfileId) of a voice registered on the MyLab TTS Voices page.

Authentication

Authentication is a bearer API key, sent as shown under Connect. API keys do not expire; contact MyLab to revoke one.

Connect

wss://api-lab.myorder.ai/api/tts/stream?voiceProfileId=<voice ID>

The request is validated before the WebSocket upgrade. A rejection is an HTTP response with the body {"error": {"code": "...", "message": "..."}}; no WebSocket is opened. WebSocket libraries surface it as a connection failure that includes the HTTP status (ws: the unexpected-response event; websockets: InvalidStatus). Browser clients receive only close code 1006.

Session lifecycle

The server sends no close reason; the close code identifies the cause:

Reconnect using exponential backoff with jitter, from 500 ms up to 30 seconds. Apply the same backoff to 429, 500 and 503 responses on connection.

Audio format

audio frames are binary WebSocket frames of raw PCM, 16-bit signed little-endian, mono, 48 kHz: no header, no container; 1 ms of audio is 96 bytes. Frames are delivered in order and vary in size.

Limits

Limits are measured in Unicode characters; each Thai vowel and tone mark counts as one. Send one or two sentences per request and split longer text at sentence boundaries. Open several sessions for parallel synthesis.

Billing

Usage is metered per request on the audio generated, in whole seconds.

Quickstart

JavaScript (Node.js, ws):

import WebSocket from 'ws';

const ws = new WebSocket(
  `wss://api-lab.myorder.ai/api/tts/stream?voiceProfileId=${VOICE_ID}`,
  { headers: { Authorization: `Bearer ${API_KEY}` } },
);
const pcm = [];

ws.on('message', (data, isBinary) => {
  if (isBinary) return pcm.push(data);
  const frame = JSON.parse(data);
  if (frame.type === 'session-ready') {
    ws.send(JSON.stringify({ type: 'synthesize', text: 'Hello, thank you for calling.' }));
  } else if (frame.type === 'done') {
    ws.close(1000);
  }
});

Python (websockets 14 or later):

import asyncio, json, websockets

async def main():
    url = f"wss://api-lab.myorder.ai/api/tts/stream?voiceProfileId={VOICE_ID}"
    async with websockets.connect(url, additional_headers={"Authorization": f"Bearer {API_KEY}"}) as ws:
        pcm = bytearray()
        async for message in ws:
            if isinstance(message, bytes):
                pcm += message
                continue
            frame = json.loads(message)
            if frame["type"] == "session-ready":
                await ws.send(json.dumps({"type": "synthesize", "text": "Hello, thank you for calling."}))
            elif frame["type"] == "done":
                break

asyncio.run(main())