OpenAI API text to speech endpoints#

API Reference#

OpenVINO Model Server includes now the audio/speech endpoint using OpenAI API. It is used to execute text to speech task with OpenVINO GenAI pipeline. Please see the OpenAI API Create Speech Reference for more information on the API. The endpoint is exposed via a path:

http://server_name:port/v3/audio/speech

Request body must be in JSON format, and the request must have Content-Type: application/json header.

Qwen3 TTS compatibility#

OVMS supports the Qwen3 TTS variants that fit the OpenAI-compatible request schema used by /v1/audio/speech:

  • Qwen3-TTS-*-CustomVoice: supported via voice and optional instructions

  • Qwen3-TTS-*-VoiceDesign: supported via instructions

Qwen3-TTS-*-Base is not currently supported through this endpoint, because it requires reference-audio inputs such as ref_audio and optionally ref_text, which are outside the standard OpenAI audio/speech request shape.

Example request#

curl http://localhost:8000/v3/audio/speech \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Kokoro-82M-int8-ov",
    "voice": "af_alloy",
    "input": "The quick brown fox jumped over the lazy dog."
  }' \
  -o speech.wav

Example response#

speech.wav - audio file in wav format.

Language handling#

For Kokoro-based speech generation, OVMS forwards the optional language field to the OpenVINO GenAI speech generation pipeline. The currently supported GenAI phonemization values are:

  • en-us

  • en-gb

  • es

  • fr-fr

  • it

  • pt-br

  • hi

If language is omitted, OVMS defaults to en-us. For non-English text, set language explicitly. Voice selection does not infer the phonemization language.

Request#

Param

OpenVINO Model Server

OpenAI /audio//speech API

Type

Description

model

✅

✅

string (required)

Name of the model to use. Name assigned to a MediaPipe graph configured to schedule generation using desired embedding model. Note: This can also be omitted to fall back to URI based routing. Read more on routing topic TODO

input

✅

✅

string (required)

The text to generate audio for.

voice

✅

✅

string

The voice to use when generating the audio. OVMS first checks graph-configured or model-local voice embeddings, then forwards the value as a model-native speaker name when the backend supports it. If not provided, model default voice will be used.

language

✅

❌

string

Optional phonemization language passed to the OpenVINO GenAI speech generation pipeline. Supported values for Kokoro are en-us, en-gb, es, fr-fr, it, pt-br, hi. Defaults to en-us.

instructions

✅

✅

string

Additional generation instructions forwarded to the OpenVINO GenAI speech generation pipeline for models that support prompt-based voice control.

response_format

❌

✅

string

The format to audio in.

speed

❌

✅

number

The speed of the generated audio.

stream_format

❌

✅

string

The format to stream the audio in.

Error handling#

Endpoint can raise an error related to incorrect request in the following conditions:

  • Incorrect format of any of the fields based on the schema

References#

End to end demo with speech generation endpoint

Code snippets

Speech Generation calculator configuration and limitations