doubao-voice-clone

name	doubao-voice-clone
description	Clone voices using Volcengine Doubao (豆包) Voice Cloning API — upload audio, train, check status, then synthesize with cloned voice
metadata	{"openclaw":{"emoji":"🎙️","requires":{"bins":"[Truncated]","env":"[Truncated]"},"primaryEnv":"VOLCENGINE_TTS_TOKEN"}}

Doubao Voice Clone (豆包声音复刻)

Clone any voice from an audio sample using Volcengine Doubao Voice Cloning API. After training, use the cloned voice for text-to-speech synthesis.

Prerequisites

1. Activate the 声音复刻 service in 豆包语音控制台
2. Order a speaker_id: Go to console → 声音复刻 → 下单 to get a speaker_id (format: S_xxxxxxx)
3. Set environment variables: VOLCENGINE_TTS_APPID, VOLCENGINE_TTS_TOKEN

Workflow

1. Upload Audio → Train a voice clone from an audio sample
2. Check Status → Poll until training completes
3. Synthesize → Use the cloned speaker_id as voice_type in TTS

APIs

1. Upload & Train

Endpoint: POST https://openspeech.bytedance.com/api/v1/mega_tts/audio/upload

Script: scripts/voice_clone_upload.py

Parameters:

Param	Flag	Required	Default	Description
audio	--audio	Yes	—	Path to audio file (wav/mp3/ogg/m4a/aac, max 10MB)
speaker_id	--speaker-id	Yes	—	Speaker ID from console (format: S_xxxxxxx)
model_type	--model-type	No	1	Clone model: 1=ICL1.0, 2=DiT standard, 3=DiT restore, 4=ICL2.0
language	--language	No	0	0=Chinese, 1=English, 2=Japanese, 3=Spanish, 4=Indonesian, 5=Portuguese
denoise	--denoise	No	—	Enable audio denoising
resource_id	--resource-id	No	seed-icl-1.0	seed-icl-1.0 for ICL1.0, seed-icl-2.0 for ICL2.0

Examples:

# Upload audio to train voice clone (ICL 1.0)
python3 scripts/voice_clone_upload.py --audio sample.wav --speaker-id S_my_voice_01

# ICL 2.0 model
python3 scripts/voice_clone_upload.py --audio sample.mp3 --speaker-id S_my_voice_02 --model-type 4 --resource-id seed-icl-2.0

# English voice with denoising
python3 scripts/voice_clone_upload.py --audio english.wav --speaker-id S_en_voice --language 1 --denoise

# DiT restore model (preserves accent and speaking style)
python3 scripts/voice_clone_upload.py --audio sample.wav --speaker-id S_dit_voice --model-type 3

Success output:

上传成功 | speaker_id: S_my_voice_01

2. Check Training Status

Endpoint: POST https://openspeech.bytedance.com/api/v1/mega_tts/status

Script: scripts/voice_clone_status.py

Parameters:

Param	Flag	Required	Description
speaker_id	--speaker-id	Yes	Speaker ID to check
resource_id	--resource-id	No	seed-icl-1.0 (default) or seed-icl-2.0

Example:

python3 scripts/voice_clone_status.py --speaker-id S_my_voice_01

Status Codes:

Code	Status	Action
0	Not started	Upload audio first
1	Training	Wait and re-check
2	Success	Ready for synthesis
3	Failed	Check error and retry

3. Synthesize with Cloned Voice

After training completes (status=2), use the speaker_id as voice_type in the TTS API:

# Using the doubao-tts skill's script:
python3 ../doubao-tts/scripts/tts_synthesize.py --text "你好" --voice S_my_voice_01 --cluster volcano_icl

# Or directly with this skill's synthesize script:
python3 scripts/voice_clone_synthesize.py --text "你好世界" --speaker-id S_my_voice_01 --output cloned.mp3

Important cluster mapping for cloned voices:

model_type	Cluster (字符版)	Cluster (并发版)
1, 2, 3 (ICL 1.0)	volcano_icl	volcano_icl_concurr
4 (ICL 2.0)	volcano_icl	volcano_icl_concurr

Model Types

model_type	Name	Description
1	ICL 1.0	Standard voice cloning, good quality
2	DiT Standard	Tone only, doesn't preserve speaking style
3	DiT Restore	Preserves accent, speed, and speaking style
4	ICL 2.0	Latest model, best quality

Audio Requirements

• Formats: wav, mp3, ogg, m4a, aac, pcm (pcm must be 24kHz mono)
• Max size: 10MB per file
• Best practice: 10-30 seconds of clear speech, minimal background noise
• Upload limit: 10 uploads per speaker_id

Error Codes

Code	Description
0	Success
1001	Bad request parameters
1101	Audio upload failed
1102	ASR transcription failed
1103	Voiceprint detection failed
1104	Voiceprint too similar to celebrity
1106	Speaker ID already exists
1107	Speaker ID not found
1108	Audio transcoding failed
1109	Audio-text mismatch too large
1111	No speech detected in audio
1112	Signal-to-noise ratio too low
1122	No human voice detected
1123	Upload limit reached (max 10 per speaker)

Environment Variables

Variable	Required	Description
VOLCENGINE_TTS_APPID	Yes	Application ID
VOLCENGINE_TTS_TOKEN	Yes	Access Token
VOLCENGINE_TTS_CLUSTER	No	Default cluster (default: volcano_icl)

References

• 声音复刻 API-V1 文档
• 声音复刻 API-V3 文档 (推荐)
• 声音复刻下单指南
• 声音复刻最佳实践
• 音色管理 API