doubao-tts

豆包 TTS 文字转语音（火山引擎）

Security Scan

The skill's requirements and runtime instructions are consistent with a Volcengine/豆包 TTS integration; it asks the user to store a service token in a local config file and then sends text to the provider to synthesize audio.

Purpose & Capability

Name/description match the actions in SKILL.md: the skill asks for Volcengine (火山引擎) AppID/Access Token/Cluster, builds a TTS request, posts it to a Bytedance/Volcengine endpoint, decodes returned audio and plays/saves it. All requested items (local config file, jq/curl/base64 usage) are proportional to a TTS CLI workflow.

Instruction Scope

Instructions only read a user config file under ~/.openclaw and the text to synthesize, then send that text to openspeech.bytedance.com. This is within expected scope for a TTS skill, but there are implementation issues worth noting: the constructed JSON uses a literal "access_token" string instead of the $access_token variable and the Authorization header uses "Bearer;" (semicolon) rather than the standard "Bearer <token>" — these are likely bugs that would break requests but are not evidence of malicious intent. The script also hardcodes a user uid value; that is unusual but not dangerous by itself.

Install Mechanism

Instruction-only skill with no install spec or external downloads. README lists dependencies (jq, curl, base64, afplay/paplay) which are reasonable and expected for the provided shell workflow.

Credentials

No environment variables are requested by the registry metadata. The skill asks the user to store AppID and Access Token in a local config file (~/.openclaw/doubao-tts-config.json) — this is appropriate for an API-based TTS integration, but storing long-lived tokens in plaintext requires caution (file permissions, token scope/rotation).

Persistence & Privilege

The skill does not request permanent/always-on presence and does not modify other skills or system-wide settings. It only reads its own config file and writes temporary audio output files.

Assessment

This skill appears to be what it says: a Volcengine (豆包) TTS helper that sends text to Bytedance's TTS endpoint and returns audio. Before installing: 1) Be aware that any text you submit will be sent to Volcengine (their servers see the text). 2) Store the Access Token in the recommended config file and restrict its permissions (e.g., chmod 600 ~/.openclaw/doubao-tts-config.json). 3) The SKILL.md contains small bugs (the payload sets "token": "access_token" literally and the Authorization header uses "Bearer;"), so requests may fail; if you rely on this skill, review/fix those lines to use the actual $access_token and a proper "Authorization: Bearer $access_token" header. 4) The source/homepage is not provided — if you require provenance, ask the publisher for the repository or verify the code before use. 5) If you have privacy or compliance concerns about sending text to an external provider, do not install or use this skill.

豆包 TTS 技能

用火山引擎豆包 TTS 把文字转成语音，支持多种音色、情感、语速调节。

---

配置步骤

1. 申请火山引擎凭证

1. 访问 https://www.volcengine.com/
2. 注册/登录账号
3. 进入控制台 → 语音技术 → 语音合成
4. 创建应用，获取以下信息：
   • AppID
   • Access Token
   • Cluster（集群 ID）

2. 创建配置文件

创建 ~/.openclaw/doubao-tts-config.json：

{
  "appid": "你的 AppID",
  "access_token": "你的 Access Token",
  "cluster": "你的 Cluster ID",
  "voice_type": "BV700_streaming",
  "emotion": "pleased"
}

配置说明

字段	必填	说明	示例
appid	是	火山引擎 AppID	1234567890
access_token	是	访问令牌	xxxxx
cluster	是	集群 ID	volcano_tts
voice_type	否	默认音色	BV700_streaming
emotion	否	默认情感	pleased

---

音色推荐

常用音色

音色名称	voice_type	特点
灿灿	BV700_streaming	22 种情感，最全能（默认）
擎苍	BV701_streaming	旁白模式，适合读文章
通用女声	BV001_streaming	日常对话
通用男声	BV002_streaming	基础男声
甜美小源	BV405_streaming	智能助手风格
知性女声	BV009_streaming	专业严肃

情感/风格

情感	值	说明
愉悦	pleased	默认，友好语气
开心	happy	欢快
悲伤	sad	低沉
客服	customer_service	专业客服
讲故事	storytelling	叙述风格
安慰鼓励	comfort	温暖安慰

---

使用方式

基础用法

朗读 "今天天气不错"
帮我把这段话转成语音：你好，欢迎使用豆包 TTS
用声音读出来：欢迎回家

指定音色

用男声读 "你好"
用灿灿的声音读这段文字
换一个女声读

自动情感判断（推荐）

技能会自动根据内容选择情感，无需手动指定：

内容特征	自动情感	示例
夸奖、好事、成功	happy（开心）	"太棒了""搞定了"
道歉、失误、遗憾	sad（悲伤）	"对不起""弄错了"
安慰、陪伴、鼓励	pleased（愉悦）	"别难过""我陪着你"
惊讶、意外	surprise（惊讶）	"哇""真的吗"
特别悲伤	tear（哭腔）	"呜呜""哭了"
日常对话	pleased（愉悦）	默认友好语气

注意：灿灿 (BV700_streaming) 支持的情感有限，以上已做适配。

# 自动判断，无需指定情感
朗读 "吴，今天天气不错"  → 自动用 pleased
朗读 "太棒了，完成了！"  → 自动用 happy
朗读 "吴对不起，我弄错了" → 自动用 sorry

调节参数

朗读 --speed 1.2 "加快语速"
朗读 --volume 0.8 "降低音量"
朗读 --pitch 1.1 "提高音调"

保存文件

朗读 --save ~/Desktop/output.mp3 "保存到这里"

---

核心逻辑

1. 读取配置

config_file="$HOME/.openclaw/doubao-tts-config.json"

if [ ! -f "$config_file" ]; then
    echo "❌ 配置文件不存在，请先创建 $config_file"
    echo "参考：https://www.volcengine.com/docs/6561/97465"
    return 1
fi

appid=$(cat "$config_file" | jq -r '.appid')
access_token=$(cat "$config_file" | jq -r '.access_token')
cluster=$(cat "$config_file" | jq -r '.cluster')
voice_type=$(cat "$config_file" | jq -r '.voice_type // "BV700_streaming"')
emotion=$(cat "$config_file" | jq -r '.emotion // "pleased"')

2. 自动情感判断

# 根据内容自动选择情感
auto_select_emotion() {
    local text="$1"
    local text_lower=$(echo "$text" | tr '[:upper:]' '[:lower:]')
    
    # 开心/成功场景
    if echo "$text_lower" | grep -qE "(太棒了 | 太好了 | 成功了 | 搞定 | 完成 | 厉害 | 优秀 | 棒|赞|赢|好开心|好兴奋)"; then
        echo "happy"
        return
    fi
    
    # 道歉/失误场景
    if echo "$text_lower" | grep -qE "(对不起 | 抱歉 | 弄错了 | 失误 | 遗憾 | 不好意思 | 我的错)"; then
        echo "sorry"
        return
    fi
    
    # 安慰/陪伴场景
    if echo "$text_lower" | grep -qE "(别难过 | 别伤心 | 我陪着你 | 有我在 | 没事的 | 会好的 | 加油 | 相信你)"; then
        echo "comfort"
        return
    fi
    
    # 讲故事/长文场景（超过 100 字或明显叙述）
    if [ ${#text} -gt 100 ] || echo "$text_lower" | grep -qE "(从前 | 故事 | 很久以前 | 话说)"; then
        echo "storytelling"
        return
    fi
    
    # 正式工作场景
    if echo "$text_lower" | grep -qE "(汇报 | 总结 | 数据 | 统计 | 分析 | 报告 | 清单 | 流程)"; then
        echo "professional"
        return
    fi
    
    # 默认愉悦
    echo "pleased"
}

emotion=$(auto_select_emotion "$text")

3. 构建请求

# 生成唯一请求 ID
reqid=$(uuidgen 2>/dev/null || echo "tts-$(date +%s)-$$")

# 默认参数
speed_ratio=${speed_ratio:-1.0}
volume_ratio=${volume_ratio:-1.0}
pitch_ratio=${pitch_ratio:-1.0}

# 构建 JSON（包含情感参数）
json_payload=$(cat <<EOF
{
  "app": {
    "appid": "$appid",
    "token": "access_token",
    "cluster": "$cluster"
  },
  "user": {
    "uid": "388808087185088"
  },
  "audio": {
    "voice_type": "$voice_type",
    "encoding": "mp3",
    "speed_ratio": $speed_ratio,
    "volume_ratio": $volume_ratio,
    "pitch_ratio": $pitch_ratio,
    "emotion": "$emotion"
  },
  "request": {
    "reqid": "$reqid",
    "text": "$text",
    "text_type": "plain",
    "operation": "query",
    "with_frontend": 1,
    "frontend_type": "unitTson"
  }
}
EOF
)

4. 发送请求

api_url="https://openspeech.bytedance.com/api/v1/tts"

response=$(curl -s -X POST "$api_url" \
  -H "Authorization: Bearer;$access_token" \
  -H "Content-Type: application/json" \
  -d "$json_payload")

# 检查错误
error_code=$(echo "$response" | jq -r '.code // 0')
if [ "$error_code" != "0" ] && [ "$error_code" != "null" ]; then
    error_msg=$(echo "$response" | jq -r '.message // "未知错误"')
    echo "❌ TTS 请求失败：$error_msg"
    return 1
fi

5. 处理响应

# 提取 base64 音频数据
audio_base64=$(echo "$response" | jq -r '.data')

if [ -z "$audio_base64" ] || [ "$audio_base64" = "null" ]; then
    echo "❌ 未获取到音频数据"
    return 1
fi

# 解码保存
output_file="${save_path:-/tmp/doubao_tts_$(date +%s).mp3}"
echo "$audio_base64" | base64 -d > "$output_file"

echo "✅ 语音生成成功：$output_file"

6. 播放音频

# Mac
if command -v afplay &> /dev/null; then
    afplay "$output_file"
# Linux
elif command -v paplay &> /dev/null; then
    paplay "$output_file"
# Windows
elif command -v powershell &> /dev/null; then
    powershell -c "(New-Object Media.SoundPlayer '$output_file').PlaySync()"
else
    echo "⚠️ 未找到播放器，音频已保存：$output_file"
fi

---

错误处理

错误	原因	解决方案
配置文件不存在	未创建 config.json	按步骤创建配置文件
401 Unauthorized	Token 过期或错误	重新获取 Access Token
403 Forbidden	AppID 无权限	检查应用配置
400 Bad Request	参数错误	检查 voice_type 是否有效
网络超时	网络问题	检查网络连接
无音频数据	API 返回异常	查看完整响应内容

---

完整示例脚本

#!/bin/bash
# doubao-tts.sh - 豆包 TTS 命令行工具

set -e

config_file="$HOME/.openclaw/doubao-tts-config.json"
text="$1"
save_path="$2"

# 检查配置
if [ ! -f "$config_file" ]; then
    echo "❌ 配置文件不存在：$config_file"
    exit 1
fi

# 读取配置
appid=$(cat "$config_file" | jq -r '.appid')
access_token=$(cat "$config_file" | jq -r '.access_token')
cluster=$(cat "$config_file" | jq -r '.cluster')
voice_type=$(cat "$config_file" | jq -r '.voice_type // "BV700_streaming"')

# 生成请求
reqid=$(uuidgen 2>/dev/null || echo "tts-$(date +%s)-$$")
json_payload=$(cat <<EOF
{
  "app": {"appid": "$appid", "token": "access_token", "cluster": "$cluster"},
  "user": {"uid": "388808087185088"},
  "audio": {"voice_type": "$voice_type", "encoding": "mp3"},
  "request": {"reqid": "$reqid", "text": "$text", "text_type": "plain", "operation": "query", "with_frontend": 1}
}
EOF
)

# 发送请求
response=$(curl -s -X POST "https://openspeech.bytedance.com/api/v1/tts" \
  -H "Authorization: Bearer;$access_token" \
  -H "Content-Type: application/json" \
  -d "$json_payload")

# 处理响应
audio_base64=$(echo "$response" | jq -r '.data')
if [ -z "$audio_base64" ]; then
    echo "❌ 请求失败：$response"
    exit 1
fi

output_file="${save_path:-/tmp/doubao_tts_$$.mp3}"
echo "$audio_base64" | base64 -d > "$output_file"
echo "✅ 生成成功：$output_file"

# 播放
afplay "$output_file" 2>/dev/null || echo "请手动播放：$output_file"

---

相关文件

• 配置文件：~/.openclaw/doubao-tts-config.json
• 输出目录：/tmp/doubao_tts_*.mp3 或用户指定路径
• 官方文档：https://www.volcengine.com/docs/6561/97465

---

注意事项

1. Token 有效期 — Access Token 可能过期，定期更新
2. 计费 — 火山引擎 TTS 有免费额度，超出后按量计费
3. 文本长度 — 单次请求建议不超过 500 字，长文本分段处理
4. 并发限制 — 避免短时间内大量请求