Audio Transcriptions API

Turns audio into text. Compatible with OpenAI’s POST /v1/audio/transcriptions; upload the audio file as multipart/form-data.

Family	Model	Notes
GPT transcription	`openai/gpt-transcribe`, `openai/gpt-4o-mini-transcribe`, `openai/gpt-4o-transcribe-diarize`	See GPT transcription
ElevenLabs	`elevenlabs/scribe_v2`	Supports word-level timestamps and speaker diarization; see ElevenLabs

Quickstart

base_urlhttps://api.ofox.io/v1

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

Model IDopenai/gpt-transcribe

Python

transcribe.py


import os
from openai import OpenAI
 
client = OpenAI(base_url="https://api.ofox.io/v1", api_key=os.environ["OFOX_API_KEY"])
 
with open("meeting.mp3", "rb") as f:
    result = client.audio.transcriptions.create(
        model="openai/gpt-transcribe",
        file=f,
        language="zh",
        prompt="以下是普通话的句子，请使用简体中文。",
    )
print(result.text)

Request parameters

filerequiredfile

The audio file, e.g. mp3 or wav. A non-audio file returns 400.

modelrequiredstring

Model ID; see the table above.

languagestring

The language of the audio, e.g. zh or en. Detected automatically if omitted.

promptstring

Prompt text for guiding word choice or proper nouns; applies to GPT transcription models. When transcribing Mandarin, pass language: "zh" together with the prompt “以下是普通话的句子，请使用简体中文。” (“The following are sentences in Mandarin; please use Simplified Chinese.”), otherwise the output may come back in Traditional Chinese characters.

response_formatstring· default json· valuesjsonverbose_json

Only json and verbose_json are supported; text, srt and vtt return 400. verbose_json applies only to ElevenLabs Scribe.

provider_optionsstring

Provider-specific parameters, sent as a JSON string in a form field, e.g. {"elevenlabs": {"diarize": true}}. See each family page.

Response


{
  "text": "今天天气很好，我们下午三点在会议室讨论新功能。",
  "usage": { "type": "duration", "seconds": 9 }
}

textstring

The transcribed text.

usageobject

The billing basis. When type is duration, billing is by audio seconds (seconds); when it is tokens, billing is by tokens (input_tokens, output_tokens).

ElevenLabs Scribe also returns words (per-word text with start and end times), language, duration and other fields; see ElevenLabs. If you need subtitle files, generate them yourself from the timestamps in words.

Common errors

HTTP	Error message	Cause	Fix
400	`Unsupported response_format 'srt'. Only 'json' and 'verbose_json' are supported on this gateway`	`text`, `srt` or `vtt` was used	Use `json`; build subtitles from the timestamps
400	`response_format 'verbose_json' is not compatible with model 'gpt-transcribe'. Use 'json' or 'text' instead.`	GPT transcription models do not support `verbose_json`	Use `json`
400	`Unsupported file format txt`	The uploaded file is not audio	Upload audio such as `mp3` or `wav`
400	`Model 'openai/gpt-5.4' is not supported on /v1/audio/transcriptions. Supported: ...`	The chosen model is not a transcription model	Use one of the models listed in the error
400	`You must provide a model parameter.`	The request was sent as JSON rather than a form	Upload the file as `multipart/form-data`

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