Loly 3.5

Text-to-Speech (WebSocket)

Streams PCM16 mono at 24 kHz, in frames of roughly 200 ms.

WS/ws/tts/stream?token=<stream_token>

On this page

Authorization
Request
Response
Errors

Code examples

import json, requests, websocket

BASE_URL = "https://studio.evomlabs.com"
API_KEY = "vc_sk_live_YOUR_KEY"
text = "Xin chào, đây là bài test giọng nói."

# 1. Ask for a short-lived token. It expires in 60 seconds, so open the
#    socket immediately afterwards.
token = requests.post(
    f"{BASE_URL}/api/v1/tts/stream-token",
    headers={"Authorization": f"Bearer {API_KEY}"},
    json={"text_length": len(text)},
).json()["data"]

# 2. Always connect to the ws_url the response returned.
ws = websocket.create_connection(f"{token['ws_url']}?token={token['stream_token']}")
ws.send(json.dumps({"type": "start", "text": text, "language": "vi"}))

pcm = bytearray()
while True:
    frame = ws.recv()
    if isinstance(frame, bytes):
        pcm.extend(frame)          # PCM16 mono at 24 kHz
        continue
    event = json.loads(frame)
    if event["type"] in ("done", "error", "cancelled"):
        break

ws.close()
print("received", len(pcm), "bytes of audio")

Authorization

Bearer credential

Send the credential in the Authorization header. Account and voice keys are accepted only where the endpoint contract permits them.

Security

The stream-token response carries the socket URL. Use that value rather than a hard-coded origin, and confirm with the operator that it resolves to a public wss endpoint before you ship an integration.

Request

Field	Type	Required	Default	Description
`type`	"start"	Yes	-	Must be start.
`text`	string	Yes	-	Must not exceed the max_length carried by the token.
`language`	string	No	`auto`	auto, or one code or full name from the 646-language catalog.
`speed`	number	No	`1.0`	Clamped to 0.5 to 1.5.

Important

The voice is locked into the stream token. The start message cannot change it.

Response

Events

// Server to client

{ "type": "start", "mode": "fast", "sample_rate": 24000,
  "format": "pcm16", "channels": 1 }

<binary frame>          // PCM16 mono, about 200 ms per frame

{ "type": "done", "mode": "fast", "audio_url": "/api/tts/file/output.wav",
  "chunks": 12, "ttfb_ms": 480, "duration_ms": 5230 }

{ "type": "error", "message": "Error description" }
{ "type": "cancelled" }

Errors

REST failures use the documented error envelope. Handle the error code instead of matching the human-readable message.

Error codes

Was this page helpful?