Fish Audio Speech-to-Text API - ASR Transcription

Transcribe audio with transcribe-1 or transcribe-1-pro, with optional timestamps and emotion cues.

Transcribe audio with transcribe-1 or transcribe-1-pro, with optional timestamps and emotion cues.

This endpoint is BETA: it accepts only multipart/form-data (file upload) or application/msgpack. JSON + Base64 audio is not supported.

Base URL

https://api.autorouter.top — Production

Authentication

BearerAuth: http (bearer) Authenticate using a Bearer Token. Format: Authorization: Bearer sk-xxxxxx

Endpoints

POST `/fishaudio/v1/asr`

Create Speech Transcription Request

Upload an audio file and receive transcribed text. Select the model via the model HTTP header.

Headers

Name	Type	Required	Description
`Authorization`	string	Yes	Request authentication. Format: `Bearer sk-xxxxxx`
`model`	string	No	ASR model. `transcribe-1` (default if omitted): general transcription; `transcribe-1-pro`: multi-speaker conversations, preserving emotion and vocal-event cues in the text. Applies to both multipart and MessagePack

Request Body

Content-Type: multipart/form-data or application/msgpack

Let your HTTP client set the multipart Content-Type and boundary. For MessagePack, set Content-Type: application/msgpack and encode audio as binary bytes.

Example:

{
  "audio": "<binary>",
  "language": "zh",
  "ignore_timestamps": false
}

Properties:

Name	Type	Required	Default	Description
`audio`	file	Yes	—	Audio file upload for multipart, or raw audio bytes for MessagePack
`language`	string \| null	No	Unset	Optional language hint, such as `en`, `zh`, or `ja`. The language is still auto-detected
`ignore_timestamps`	boolean	No	`true`	Set to `false` to request timestamped segments. Alignment adds processing time; more noticeable for audio shorter than 30 seconds

Responses

200: Transcription succeeded

Content-Type: application/json

401: Unauthorized

Content-Type: application/json

402: Insufficient balance

Content-Type: application/json

503: Service overloaded

Content-Type: application/json

The examples below use multipart. Let the HTTP client set Content-Type (including the boundary). Do not hardcode multipart/form-data without a boundary—that breaks the request. The Go example sets it via FormDataContentType(). For MessagePack, see below.

curl --request POST https://api.autorouter.top/fishaudio/v1/asr \
  --header "Authorization: Bearer sk-xxxxxx" \
  --header "model: transcribe-1-pro" \
  --form audio=@conversation.wav \
  --form ignore_timestamps=false

MessagePack Request Example

Set Content-Type: application/msgpack and encode audio as raw binary bytes.

import msgpack
import requests

url = "https://api.autorouter.top/fishaudio/v1/asr"

with open("conversation.wav", "rb") as f:
    payload = msgpack.packb({
        "audio": f.read(),
        "ignore_timestamps": False,
    })

headers = {
    "Authorization": "Bearer sk-xxxxxx",
    "model": "transcribe-1-pro",
    "Content-Type": "application/msgpack",
}

response = requests.post(url, headers=headers, data=payload)
print(response.text)

Response Example

{
  "text": "<|speaker:0|>你好。<|speaker:1|>[高兴]很开心认识你。<|speaker:0|>我也是。",
  "duration": 6.4,
  "segments": [],
  "language_code": "zh",
  "language": "Chinese"
}

Response fields:

Name	Type	Description
`text`	string	Full transcript. Pro includes inline speaker markers such as `<\|speaker:0\|>` and can include emotion and vocal-event cues such as `[高兴]` (happy) or `[laughter]`
`duration`	number	Audio duration in seconds
`segments`	object[]	Array of `{ "text", "start", "end" }` with timestamps in seconds. Empty when timestamps are skipped, alignment is unavailable, or no speech is detected
`segments[].text`	string	Segment text (excludes speaker, emotion, and event markers)
`segments[].start`	number	Segment start time in seconds
`segments[].end`	number	Segment end time in seconds
`language_code`	string \| null	Detected ISO 639-1 language code, e.g. `en` or `ja`. Use this field in application logic
`language`	string \| null	Detected language name, e.g. `English`, for display only

Multi-speaker response format

Pro returns speaker turns inside the text string, using <|speaker:N|> markers. The numeric label N identifies the speaker for the text that follows, up to the next marker. A repeated label means that speaker is speaking again. Labels apply within the recording; they are not names or identities shared across requests.

In the 200 response example above (with ignore_timestamps=true), read it as speaker 0 saying 你好。, speaker 1 saying [高兴]很开心认识你。, and speaker 0 saying 我也是。. The [高兴] cue describes the delivery of speaker 1's speech.

With ignore_timestamps=false (as in the curl example above), the same markers remain in text. The segments array contains timed speech with text, start, and end; alignment excludes speaker, emotion, and event markers. Segments are not speaker turns and do not contain a speaker_id field.

The API returns no separate speaker list or emotion field. Parse the inline markers in text when your application needs speaker turns or emotion cues.