豆包语音合成 2.0

豆包语音合成 2.0，支持情绪控制、多音色、语音指令。34 种音色可选，含 JARVIS 同款男声。

Security Scan

The skill is a plausible TTS client for 火山引擎 (openspeech.bytedance.com) and mostly behaves as expected, but there are inconsistencies (registry metadata omits required credentials) and small oddities (nonstandard 'Bearer; ' token format and configuration/file access) that warrant caution before installing.

Purpose & Capability

The name, description, SKILL.md, docs, and code all describe a TTS client calling a 火山引擎/openspeech.bytedance.com API — requiring APP_ID and Access Token is consistent with that purpose. However the skill registry metadata lists no required environment variables or primary credential, which is inconsistent with the shipped code and documentation that both require VOLCANO_APP_ID and VOLCANO_ACCESS_TOKEN.

Instruction Scope

Runtime instructions and code are limited to: reading env vars and an OpenClaw JSON config, posting JSON to the documented TTS API endpoint, parsing streamed responses, saving audio files under ~/.openclaw/tts_output, and optionally invoking local players (afplay/aplay/Windows start). The skill does not instruct collection of unrelated system data or sending data to unexpected external endpoints beyond openspeech.bytedance.com. Note: SKILL.md and docs repeatedly recommend placing credentials in ~/.openclaw/openclaw.json which the client will read.

Install Mechanism

There is no install spec (instruction-only skill), and included files are pure Python with no external arbitrary downloads. This is low risk from an installation mechanism perspective.

Credentials

The code and SKILL.md require VOLCANO_APP_ID and VOLCANO_ACCESS_TOKEN (and optionally VOLCANO_RESOURCE_ID / VOLCANO_API_URL), which are proportional to a TTS client. However the skill registry metadata declared no required env vars or primary credential — an incoherence that may mislead users. Also the docs and code use an unusual token header format 'Authorization: Bearer; {token}' (semicolon + space) — this is nonstandard and should be verified against the provider to ensure it is intentional and not a bug that might cause manual token workarounds.

Persistence & Privilege

The skill does write output to ~/.openclaw/tts_output and will read ~/.openclaw/openclaw.json; this is reasonable for a client that stores outputs and reads local configuration. The skill is not marked always:true and does not request system-wide persistent privileges beyond its own config and output directory.

What to consider before installing

Things to check before installing/use: - The skill's registry metadata did not list required environment variables, but both SKILL.md and the shipped Python code require VOLCANO_APP_ID and VOLCANO_ACCESS_TOKEN. Don't provide credentials until you confirm this is the intended provider and you trust the package source. - Verify the API endpoint and token format with the official 火山引擎 (Volcano/ByteDance) docs. The code and docs use an unusual 'Bearer; {token}' header format (semicolon + space) — confirm that's required and not a typo that will prompt manual token hacks. - Inspect the included tts_client.py locally (it is provided) before running. The code performs network calls to https://openspeech.bytedance.com and writes audio files under ~/.openclaw/tts_output and reads ~/.openclaw/openclaw.json; ensure you are comfortable with that file access. - Playback uses subprocess.run with system binaries (afplay/aplay/Windows start). Running these is typical but ensure your environment has the expected players; subprocess usage appears to use list form (no shell injection), but you should still inspect the tail of the file (the provided tts_client.py was partially truncated in the scan) to confirm no unexpected shell=True or unescaped concatenation. - If you want least privilege: create a dedicated minimal-volume API key on the provider side (if supported) and test in a sandboxed account before using production credentials. Given the above inconsistencies and the need to handle credentials, proceed with caution and verify the token/header format and the skill's origin before granting credentials or installing into a production agent.

豆包 SeedTTS 2.0 Skill

火山引擎豆包语音合成 2.0，支持 34 种音色、情绪控制、语音指令。

---

🎯 功能特性

• ✅ 34 种音色 - 中文/英文、男声/女声、通用/角色扮演
• ✅ 情绪控制 - 支持语音标签，如 [温柔地] 你好呀
• ✅ 多格式输出 - MP3/WAV，24kHz 高采样率
• ✅ 批量生成 - 一次调用生成多个音频
• ✅ 命令行工具 - 一行命令生成语音
• ✅ Python SDK - 简化调用，自动配置

---

📦 安装

clawhub install seedtts2

---

🔧 配置

1. 获取火山引擎凭证

1. 登录 火山引擎控制台
2. 进入「语音合成」→「应用管理」
3. 创建应用或选择已有应用
4. 获取 APP ID 和 Access Token

2. 配置凭证

方式一：环境变量（推荐）

export VOLCANO_APP_ID="你的 APP ID"
export VOLCANO_ACCESS_TOKEN="你的 Access Token"
export VOLCANO_RESOURCE_ID="seed-tts-2.0"  # 可选，默认值

方式二：OpenClaw 配置

在 ~/.openclaw/openclaw.json 中添加：

{
  "skills": {
    "entries": {
      "seedtts2": {
        "env": {
          "VOLCANO_APP_ID": "你的 APP ID",
          "VOLCANO_ACCESS_TOKEN": "你的 Access Token"
        }
      }
    }
  }
}

---

🚀 使用方法

命令行

# 基础用法
seedtts2 "你好，这是测试"

# 指定输出文件
seedtts2 "你好" -o test.mp3

# 指定音色
seedtts2 "你好" -s zh_male_ruyayichen_uranus_bigtts

# 合成后播放
seedtts2 "你好" -p

# 列出可用音色
seedtts2 --list

Python SDK

from tts_client import SeedTTS2

# 初始化（自动从环境变量读取配置）
tts = SeedTTS2()

# 基础合成
tts.say("你好，这是测试")

# 指定音色（JARVIS 官方音色）
tts.say("你好", speaker="zh_male_ruyayichen_uranus_bigtts")

# 保存到指定文件
tts.say("你好", output="test.mp3")

# 合成并播放
tts.say_and_play("你好")

# 批量生成
tts.batch_generate([
    {"text": "第一句", "speaker": "zh_male_ruyayichen_uranus_bigtts"},
    {"text": "第二句", "speaker": "zh_female_vv_uranus_bigtts"},
], output_dir="./output")

快捷函数

from tts_client import say, say_and_play, batch_generate

# 一句话生成
say("你好")

# 生成并播放
say_and_play("你好")

# 批量生成
batch_generate(["第一句", "第二句"], output_dir="./output")

---

🎵 音色列表

推荐音色

音色	Voice Type	场景
儒雅逸辰 2.0	zh_male_ruyayichen_uranus_bigtts	JARVIS 官方，成熟稳重
Vivi 2.0	zh_female_vv_uranus_bigtts	表现力强，情感丰富
甜美小源 2.0	zh_female_tianmeixiaoyuan_uranus_bigtts	甜美温柔，情感陪伴
云舟	zh_male_m191_uranus_bigtts	沉稳大气
Tim	en_male_tim_uranus_bigtts	标准美式英语
Dacey	en_female_dacey_uranus_bigtts	标准英式英语

完整音色库

运行以下命令查看完整音色列表：

seedtts2 --list

或参考文档：seedtts2-voice-library.md

---

🎭 语音指令

支持在文本中使用语音标签控制情绪和语气：

【格式】[表情/心理/肢体动作] 台词

【示例】
[温柔地] 你好呀
[开心地] 太好啦！
[严肃地] 请注意，这是重要通知
[旁白，语调平静] 故事从这里开始

---

📁 目录结构

seedtts2/
├── SKILL.md              # 本文件
├── bin/
│   └── seedtts2          # 命令行工具
├── examples/
│   ├── basic_usage.py    # 基础示例
│   ├── batch_generate.py # 批量生成
│   └── emotion_control.py # 情绪控制
├── docs/
│   ├── voice-library.md  # 完整音色库
│   └── troubleshooting.md # 故障排查
└── tts_client.py         # Python SDK

---

⚠️ 故障排查

问题 1：缺少必要配置

错误信息：

ValueError: 缺少必要配置：APP_ID, ACCESS_TOKEN

解决方案：

• 检查环境变量是否设置：echo $VOLCANO_APP_ID
• 或检查 openclaw.json 配置是否正确

问题 2：认证失败（401）

错误信息：

请求失败：{"message": "Unauthorized"}

解决方案：

• 检查 Access Token 是否正确
• 确认 Token 格式为 Bearer; {token}（注意空格）
• 在火山引擎控制台确认应用状态

问题 3：音色不可用（403）

错误信息：

请求失败：{"message": "Forbidden"}

解决方案：

• 该音色可能需要单独开通
• 登录火山引擎控制台 → 「音色管理」
• 确认音色状态为"已开通"

问题 4：音频无法播放

解决方案：

• macOS：使用 afplay 播放
• Linux：安装 aplay（alsa-utils 包）
• Windows：使用默认播放器

---

📚 相关文档

• 部署手册 - 完整配置指南
• 音色库 - 34 种音色详情
• 复盘报告 - 经验教训总结

---

🙏 致谢

• 火山引擎豆包团队提供强大的语音合成 API
• OpenClaw 社区提供 Skill 开发框架

---

📄 许可证

MIT License

---

维护者: JARVIS
最后更新: 2026-03-24
版本: 1.0.0