---

name: bailian-tts description: Generate speech audio with 阿里云百炼 TTS via the bailian-cli npm package. Use when users ask to convert text to voice, choose voices/languages, batch-generate narration, or troubleshoot 百炼 TTS setup (npm install, API key, region, output format/path).

Bailian TTS

Use this skill to produce speech audio from text with @hackerpl/bailian-cli.

Quick Workflow

1. Check CLI availability.
2. Check BAILIAN_API_KEY.
3. If key is missing, guide user to prepare one in 阿里云百炼.
4. Run bailian tts with requested voice/language/style.
5. Return output path (or base64 when requested). Default audio output directory: ~/.openclaw/media/audio.

1) Check and install CLI

Run:

npm ls -g --depth=0 @hackerpl/bailian-cli

If missing, install:

npm i -g @hackerpl/bailian-cli

2) Check API key

Run:

[ -n "$BAILIAN_API_KEY" ] && echo SET || echo MISSING

If missing, ask user to create and configure key from the 阿里云百炼官方入口（控制台/平台）：

• https://bailian.console.aliyun.com/

Then set it:

export BAILIAN_API_KEY="sk-xxxx"
# optional
export BAILIAN_REGION="beijing"   # or singapore

3) Generate speech (bailian-cli)

Basic (default to OpenClaw media folder):

mkdir -p ~/.openclaw/media/audio
bailian tts -t "你好，欢迎使用百炼 TTS 服务" -o url -d ~/.openclaw/media/audio

Specify voice/language:

mkdir -p ~/.openclaw/media/audio
bailian tts -t "Hello world" -v "Ethan" -l "English" -o url -d ~/.openclaw/media/audio

With style instruction:

mkdir -p ~/.openclaw/media/audio
bailian tts -t "春眠不觉晓" -v "Serena" -i "用温柔缓慢的语调朗读" -o url -d ~/.openclaw/media/audio

Return base64 instead of file:

bailian tts -t "测试" -o data

Custom output directory:

bailian tts -t "你好世界" -d "./audio"

4) Voices and languages

• For live lookup:

bailian tts --list-voices

• Full built-in table: read references/voices-and-languages.md.

5) Operational notes

• Keep text length within CLI limit (600 chars per request).
• Default output directory: ~/.openclaw/media/audio (create with mkdir -p ~/.openclaw/media/audio).
• Prefer -o url for file delivery, -o data for programmatic pipelines.
• Respect user privacy: do not upload sensitive text externally without explicit confirmation.
• If generation fails, check in order: API key → region → network → voice/language spelling.
• When producing many clips, keep a consistent voice and instruction style for tonal continuity.