Voice and transcription

Text to speech

POST /v1/audio/speech

Returns the audio file directly.

from openai import OpenAI

client = OpenAI(base_url="http://localhost:7860/v1", api_key="not-needed")
with client.audio.speech.with_streaming_response.create(
    model="auto", voice="default", input="Hello there! This was made on my own computer."
) as response:
    response.stream_to_file("hello.mp3")

Field	Notes
input	Required. The text to read.
model	auto, a model id or a Hugging Face repo id.
voice	For Bark models, a speaker preset such as v2/en_speaker_6. Other voice models have a single voice and ignore it. The OpenAI SDK requires the field, so send any value, such as default.
response_format	mp3 (default), wav, flac or ogg. Other formats fall back to mp3.
speed	Default 1.0.

Transcription

POST /v1/audio/transcriptions

Multipart form upload.

result = client.audio.transcriptions.create(model="whisper-1", file=open("meeting.mp3", "rb"))
print(result.text)

whisper-1 (OpenAI’s model name) is treated like auto, so existing code works unchanged.

Field	Notes
file	Required. The recording.
model	auto, whisper-1, a model id or a Hugging Face repo id.
language	Optional language code, such as en or fr. Detected automatically if left out.
response_format	json (default, {"text": ...}), text, verbose_json (with timed segments), srt or vtt.

Timed segments, and so srt and vtt, come from Whisper models.

Translation

POST /v1/audio/translations

Same as transcription, but the text comes out in English whatever language is spoken. Use a Whisper model.

curl http://localhost:7860/v1/audio/translations -F file=@interview-es.mp3 -F response_format=text