Audio Speech API

Turns text into speech. Compatible with OpenAI’s POST /v1/audio/speech; the response body is the audio file itself (binary), not JSON.

Family	Model	Notes
ElevenLabs	`elevenlabs/eleven_v4`	See ElevenLabs
Gemini TTS	`google/gemini-3.8-flash-tts` and others	See Gemini TTS

Accepted values for voice, response_format and other parameters vary by model; the family pages are authoritative.

Quickstart

base_urlhttps://api.ofox.io/v1

Endpointhttps://api.ofox.io/v1/audio/speech

Model IDgoogle/gemini-3.8-flash-tts

Python

speech.py


import os
from openai import OpenAI
 
client = OpenAI(base_url="https://api.ofox.io/v1", api_key=os.environ["OFOX_API_KEY"])
 
with client.audio.speech.with_streaming_response.create(
    model="google/gemini-3.8-flash-tts",
    voice="Kore",
    input="Hello, and welcome to OfoxAI text to speech.",
    response_format="wav",
) as resp:
    resp.stream_to_file("speech.wav")

Request parameters

modelrequiredstring

Model ID; see the table above.

inputrequiredstring

The text to synthesize. An empty value returns 400.

voicestring

The voice. Required for Gemini TTS, where it takes a Google voice name (e.g. Kore); for ElevenLabs it takes a voice ID (e.g. 21m00Tcm4TlvDq8ikWAM). OpenAI voice names (e.g. alloy) do not work with either family.

response_formatstring

Audio format. Accepted values vary by model: Gemini TTS supports wav and pcm; ElevenLabs supports mp3, opus, wav and pcm. An unsupported format returns 400.

instructionsstring

Tone and style instructions, such as “read this in a cheerful tone”. Applies to Gemini TTS.

language_codestring

Language code, such as cmn-CN (Gemini TTS) or zh (ElevenLabs). If omitted, the model infers the language from the text.

stream_formatstring· valuessse

With sse, audio chunks are streamed as Server-Sent Events. Applies to Gemini TTS.

provider_optionsobject

Provider-specific parameters, passed through to the provider, e.g. ElevenLabs’ {"elevenlabs": {"voice_settings": {...}}}. See each family page.

Response

On success, HTTP 200 with the audio file as the response body; Content-Type matches the format:

response_format	Content-Type
`mp3`	`audio/mpeg`
`wav`	`audio/wav`
`opus`	`audio/ogg`
`pcm`	`audio/L16;rate=<sample rate>;channels=1`, headerless 16-bit PCM

Write the response body straight to a file. Errors are returned as JSON with Content-Type application/json, so check the status code first to avoid saving an error message as an audio file.

Common errors

HTTP	Error message	Cause	Fix
400	`model and input are required.`	`model` is missing, or `input` is empty	Supply the missing parameters
400	`voice is required for speech completion`	`voice` not sent for Gemini TTS	Pass a voice name, such as `Kore`
400	`unsupported Gemini response_format: mp3`	The model does not support the chosen format	Use a format listed on the family page
400	`Request contains an invalid argument.`	The Gemini TTS voice name does not exist	Use a valid voice name
404	`A voice with voice_id 'alloy' was not found.`	The ElevenLabs voice ID does not exist	Use a valid voice ID
400 `endpoint_not_supported`	`Model 'openai/gpt-5.4' does not support the /v1/audio/speech endpoint on this platform.`	The chosen model is not a speech synthesis model	Use a model from the table above

For the full list of error codes, see Error handling.