doubao-tts

name	doubao-tts
description	Text-to-Speech synthesis using Volcengine Doubao (豆包) Speech API with 2.0 voice instruction support
metadata	{"openclaw":{"emoji":"🗣️","requires":{"bins":"[Truncated]","env":"[Truncated]"},"primaryEnv":"VOLCENGINE_TTS_TOKEN"}}

Doubao TTS (豆包语音合成)

Synthesize natural speech from text using Volcengine Doubao Speech API. Supports 2.0 big-model voices with voice instructions, emotion control, context reference, and voice tags.

Workflow

1. Prepare text — plain text, with optional voice instructions [#指令] or context [#上文]
2. Run script — call tts_synthesize.py with text + voice + audio params
3. Get audio — output file saved to specified path (mp3/wav/pcm/ogg_opus)

Quick Start

# Basic synthesis (default: 可爱女生 voice, mp3)
python3 scripts/tts_synthesize.py --text "你好，今天天气真不错！" --output hello.mp3

# With voice instruction (2.0 feature)
python3 scripts/tts_synthesize.py --text "[#用开心兴奋的语气说]太棒了！我们成功了！" --output happy.mp3

# Male voice, slower speed
python3 scripts/tts_synthesize.py --text "这是一段测试语音。" --voice zh_male_vv_young_bigtts --speed 0.8 --output test.mp3

API

Synthesize

Endpoint: POST https://openspeech.bytedance.com/api/v1/tts

Auth: Authorization: Bearer;{VOLCENGINE_TTS_TOKEN}

Script: scripts/tts_synthesize.py

Parameters:

Param	Flag	Required	Default	Description
text	--text	Yes	—	Text to synthesize (max 1024 bytes UTF-8). Supports voice instructions.
voice	--voice	No	zh_female_vv_uranus_bigtts	Voice type ID or built-in name
output	--output	No	output.mp3	Output audio file path
encoding	--encoding	No	mp3	Audio format: mp3, wav, pcm, ogg_opus
speed	--speed	No	1.0	Speech rate (0.2–3.0)
volume	--volume	No	1.0	Volume (0.1–3.0)
pitch	--pitch	No	1.0	Pitch (0.1–3.0)
emotion	--emotion	No	—	Emotion style (e.g. happy, sad, angry)
language	--language	No	—	Language hint: zh-cn, en, ja, crosslingual
cluster	--cluster	No	volcano_tts	API cluster (from env VOLCENGINE_TTS_CLUSTER)

Examples:

# Default voice
python3 scripts/tts_synthesize.py --text "你好世界"

# Specify voice + format + output
python3 scripts/tts_synthesize.py --text "Hello World" --voice zh_female_vv_uranus_bigtts --encoding wav --output hello.wav

# Adjust speed and volume
python3 scripts/tts_synthesize.py --text "慢速大声播报" --speed 0.7 --volume 1.5

# With emotion (for voices that support it)
python3 scripts/tts_synthesize.py --text "这太让人难过了" --emotion sad

# English-only mode
python3 scripts/tts_synthesize.py --text "This is an English test." --language en

Response: Audio binary saved to --output path. Script prints duration and file size.

Success output:

合成成功 | 音频时长: 3200ms | reqid: uuid-here
音频已保存: output.mp3 (65,280 bytes)

Voice Types

2.0 Big-Model Voices (Recommended)

These voices support voice instructions, context reference, and voice tags.

Name	voice_type	Gender	Features
可爱女生	zh_female_vv_uranus_bigtts	Female	Voice instructions, tags
调皮公主	zh_female_vv_princess_bigtts	Female	Voice instructions, tags
爽朗少年	zh_male_vv_young_bigtts	Male	Voice instructions, tags
天才同桌	zh_male_vv_genius_bigtts	Male	Voice instructions, tags

Standard Streaming Voices

Name	voice_type	Gender
灿灿	BV700_streaming	Female
梓梓	BV406_streaming	Female
通用女声	BV001_streaming	Female
通用男声	BV002_streaming	Male
电影解说小帅	BV411_streaming	Male
电影解说小美	BV412_streaming	Female
阳光男声	BV123_streaming	Male
知性姐姐	BV034_streaming	Female

Full list: https://www.volcengine.com/docs/6561/1257544

Note: Standard voices (BV*) require volc.tts.default resource. Big-model voices (*_bigtts) require volc.megatts.default resource. Use the cluster and voices that match your account's activated service.

2.0 Features

Voice Instructions (语音指令)

Prepend [#instruction] to control emotion, dialect, tone, speed, pitch:

# Angry tone
python3 scripts/tts_synthesize.py --text "[#你得跟我互怼！用吵架的语气]你也不是什么好东西！"

# Whisper / ASMR
python3 scripts/tts_synthesize.py --text "[#用asmr悄悄话的语气]每次听到你的声音，我都觉得心里暖暖的。"

# Complex emotion
python3 scripts/tts_synthesize.py --text "[#用低沉沙哑、带着沧桑与绝望的语气说]我多想再回到从前啊。"

# Dialect
python3 scripts/tts_synthesize.py --text "[#用四川话说]巴适得板！"

Context Reference (引用上文)

Provide preceding context that the model understands but doesn't synthesize:

# Q&A context — model understands the question and speaks the answer naturally
python3 scripts/tts_synthesize.py --text "[#你怎么评价北京？]北京嘛，因为我来这是第二次了。"

# Emotional context
python3 scripts/tts_synthesize.py --text "[#是你吗？好像没怎么变啊？]你头发长了，十年了，你还好吗？"

Voice Tags (语音标签)

For 2.0 voices (可爱女生, 调皮公主, 爽朗少年, 天才同桌), add expression/action/mood tags inline:

python3 scripts/tts_synthesize.py --text "【旁白，语调惊恐】他的手触碰到对方身体时，却感觉一阵冰冷僵硬。"

python3 scripts/tts_synthesize.py --text "【怒目圆睁，大声怒吼】放肆！我是龙族的女王！"

Audio Parameters

Parameter	Range	Default	Notes
encoding	mp3, wav, pcm, ogg_opus	mp3	wav does not support streaming
speed_ratio	0.2 – 3.0	1.0	Speech rate
volume_ratio	0.1 – 3.0	1.0	Volume level
pitch_ratio	0.1 – 3.0	1.0	Pitch (standard voices only; big-model voices don't support pitch)
rate	8000, 16000, 24000	24000	Sample rate in Hz
emotion	happy, sad, angry, fear, hate, surprise, neutral	—	Only for voices with multi-emotion support
language	zh-cn, en, ja, crosslingual, etc.	auto	Explicit language hint
loudness_ratio	0.5 – 2.0	1.0	Alternative volume control for big-model voices

Error Handling

Code	Description	Action
3000	Success	Normal
3001	Invalid request	Check parameters (voice_type, cluster, auth)
3003	Concurrency limit	Retry later
3005	Server busy	Retry later
3010	Text too long	Reduce to ≤1024 bytes UTF-8
3011	Invalid text	Check text content (not empty, matches language)
3030	Timeout	Retry or shorten text

Common error messages:

• "requested resource not granted" → Voice not activated in console. Buy/order the voice pack (some are free at ¥0).
• "authenticate request: load grant: requested grant not found" → Check AppID + Token.
• "illegal input text!" → Text is empty or contains only punctuation/emoji.

Environment Variables

Variable	Required	Description
VOLCENGINE_TTS_APPID	Yes	Application ID from Volcengine console
VOLCENGINE_TTS_TOKEN	Yes	Access Token (Bearer token)
VOLCENGINE_TTS_CLUSTER	No	Cluster name (default: volcano_tts)

Get credentials: https://console.volcengine.com → 豆包语音 → FAQ Q1

References

• 豆包语音合成 2.0 能力介绍
• HTTP 接口参数说明
• 大模型音色列表
• 大模型 V1 HTTP 接口文档
• SSML 标记语言