TokenLab

Audio & Realtime

Create Speech

Generates audio from the input text

POST
/v1/audio/speech

Use GET /v1/models?recommended_for=tts to get current speech models.

Request Body

The response mode depends on the selected model and stream_format. For longer input, allow an HTTP client timeout of at least 120s.

Optional fields, voices, output formats, and SSE support depend on the selected model. Check GET /v1/models/{model} before sending them. Unsupported parameters can return 400 unsupported_parameter.

modelstringrequired

TTS model ID. Find current choices with GET /v1/models?recommended_for=tts. There is no default model.

inputstringrequired

The text to generate audio for. Maximum 4096 characters; a model can set a lower limit, see its details.

voicestring | object

Voice selector. Pass a voice name from the model's details, such as eve for grok-tts, or an object like { "id": "voice-id" } for models that accept voice IDs.

voice_idstring

Voice ID for models that list this parameter, such as speech-2.8-turbo.

instructionsstring

Optional style or delivery instructions, up to 2000 characters, for models that list this parameter.

promptstring

Optional speaking style prompt, up to 2000 characters, for models that list this parameter, such as qwen3-tts-1-7b-voicedesign.

language_codestring

Language code, such as en-US, for models that support it.

response_formatstring

Audio format. Accepted values depend on the model: for example grok-tts returns mp3 only, and qwen3-tts-flash returns wav. Check the model's details before sending a value.

stream_formatstringdefault: audio

TokenLab delivery format: audio or sse. Send sse only when the model's details list it.

speednumber

Speech speed for models that support it. TokenLab accepts 0.25 to 4.0; a model can narrow the range.

temperaturenumber

Sampling temperature from 0 to 2, for models that list this parameter.

Response

With stream_format=audio, the response contains audio bytes. Models that support stream_format=sse return a text/event-stream response; parse its events instead of saving the entire response as an audio file.

Request

curl -X POST "https://api.tokenlab.sh/v1/audio/speech" \
  -H "Authorization: Bearer sk-your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "grok-tts",
    "voice": "eve",
    "input": "Hello, welcome to TokenLab!"
  }' \
  --output speech.mp3

Voice Samples

Voices depend on the model. Read the voice values in the model details from GET /v1/models/{model} before you send one. For example, grok-tts accepts eve, ara, rex, sal, and leo.

Response example

Response

200 OK
<binary audio data>

Important fields

Content-Typestring

HTTP response media type: the audio format, or text/event-stream for SSE.

bodybinary | SSE

Audio bytes for audio; an event stream for sse. Choose the parser using Content-Type and the requested mode.

MiniMax Speech 2.8 Turbo synthesizes a script in a selected voice with a speed-focused workflow, useful for responsive spoken interfaces and narration.

{
  "model": "speech-2.8-turbo",
  "input": "Welcome to TokenLab.",
  "voice_id": "male-qn-qingse",
  "speed": 1,
  "response_format": "mp3",
  "stream_format": "audio"
}

Authorization

BearerAuth
AuthorizationBearer <token>

API Key authentication. Create or manage API keys in Dashboard > API > API Keys.

In: header

Headers

X-TokenLab-Delivery-Policy?string

Per-request Delivery policy. Overrides the API key and Workspace defaults. Auto tries TokenLab Verified first and may switch once to Official only before output, request acceptance, or persistent resource creation.

Value in

  • "auto"
  • "verified"
  • "official"

Request Body

application/json

Response

application/json

application/json

application/json

application/json

application/json

application/json