doubao-asr

name	doubao-asr
description	Transcribe audio files to text using Volcengine Doubao (豆包) Big-Model ASR 2.0 with word-level timestamps
metadata	{"openclaw":{"emoji":"👂","requires":{"bins":"[Truncated]","env":"[Truncated]"},"primaryEnv":"VOLCENGINE_TTS_TOKEN"}}

Doubao ASR (豆包录音文件识别 2.0)

Transcribe audio files to text using Volcengine Doubao Big-Model ASR. Supports word-level timestamps, punctuation, ITN (inverse text normalization), speaker diarization, and channel splitting.

Workflow

1. Submit — Upload audio (file or URL) to start transcription
2. Query — Poll for results (small files return instantly; large files are async)
3. Get Results — Full text + word-level timestamps + optional speaker info

Quick Start

# Transcribe a local audio file
python3 scripts/asr_transcribe.py --audio recording.mp3

# Transcribe from URL
python3 scripts/asr_transcribe.py --url "https://example.com/audio.wav"

# Save transcription to file
python3 scripts/asr_transcribe.py --audio meeting.mp3 --output transcript.txt

# With speaker diarization
python3 scripts/asr_transcribe.py --audio meeting.mp3 --speakers

API

Transcribe

Submit: POST https://openspeech.bytedance.com/api/v3/auc/bigmodel/submit Query: POST https://openspeech.bytedance.com/api/v3/auc/bigmodel/query

Script: scripts/asr_transcribe.py

Parameters:

Param	Flag	Required	Default	Description
audio	--audio	Yes*	—	Path to local audio file
url	--url	Yes*	—	URL to audio file
output	--output	No	—	Save transcript to text file
format	--format	No	auto-detect	Audio format: mp3, wav, ogg, m4a, flac, aac, amr
speakers	--speakers	No	false	Enable speaker diarization
channels	--channels	No	false	Enable channel splitting
words	--words	No	false	Show word-level timestamps
max-wait	--max-wait	No	120	Max seconds to wait for async tasks

*One of --audio or --url is required.

Examples:

# Basic transcription
python3 scripts/asr_transcribe.py --audio podcast.mp3

# Show word-level timestamps
python3 scripts/asr_transcribe.py --audio speech.wav --words

# Multi-speaker meeting with output file
python3 scripts/asr_transcribe.py --audio meeting.mp3 --speakers --output meeting.txt

# From URL with channel splitting
python3 scripts/asr_transcribe.py --url "https://example.com/call.wav" --channels

# Custom audio format
python3 scripts/asr_transcribe.py --audio recording.raw --format pcm

Success output:

音频时长: 5496ms
识别结果: 你好，我是豆包语音合成，今天天气真不错，一起出去走走吧。

With --words:

音频时长: 5496ms
识别结果: 你好，我是豆包语音合成，今天天气真不错，一起出去走走吧。

逐字时间戳:
  [0.19s-0.35s] 你
  [0.35s-0.71s] 好
  [0.71s-0.99s] 我
  ...

Request Parameters (Advanced)

These can be set in the request object when calling the API directly:

Parameter	Type	Default	Description
model_name	string	bigmodel	Model name
model_version	string	400	Model version
enable_itn	bool	true	Inverse text normalization (numbers → digits)
enable_punc	bool	true	Automatic punctuation
enable_ddc	bool	true	Spoken language normalization
show_utterances	bool	true	Return sentence-level segments
enable_channel_split	bool	false	Split audio channels (for stereo)
enable_speaker_info	bool	false	Speaker diarization

Auth Headers

The ASR API uses header-based auth (different from TTS Bearer token):

Header	Value
X-Api-App-Key	AppID
X-Api-Access-Key	Access Token
X-Api-Resource-Id	volc.bigasr.auc
X-Api-Request-Id	Unique task UUID
X-Api-Sequence	-1 (single request)

Response Status Codes

Code	Status	Action
20000000	Completed	Results available
20000001	Queued	Wait and re-query
20000002	Processing	Wait and re-query
Other	Failed	Check error message

Supported Audio Formats

mp3, wav, ogg, m4a, flac, aac, amr, pcm (16kHz 16bit mono)

Environment Variables

Variable	Required	Description
VOLCENGINE_TTS_APPID	Yes	Application ID
VOLCENGINE_TTS_TOKEN	Yes	Access Token

References

• 大模型录音文件识别标准版 API
• 大模型录音文件识别极速版 API
• 豆包语音产品文档