audio-summary

name: audio-summary description: Generate conversational audio summaries using text-to-speech. Supports multiple providers (Gemini TTS, OpenAI TTS, ElevenLabs) with preference order and automatic fallback. metadata: openclaw: emoji: 🔊 requires: envVars: - GEMINI_API_KEY | OPENAI_API_KEY | ELEVENLABS_API_KEY

Audio Summary Skill

Generates conversational audio summaries using multi-provider TTS.

When to Use

• User says "give me an audio summary"
• User says "audio summary please"
• User says "tell me in voice"
• User explicitly requests audio/voice response

How It Works

1. Generate summary text — Create a conversational summary (150 words max, 30-60 seconds)
2. Convert to audio — Use TTS provider (Gemini → OpenAI → ElevenLabs fallback)
3. Deliver audio — Upload via your chat platform or serve locally

Provider Preference Order

Note: Gemini outputs .wav (PCM wrapped with WAV header), not .mp3. The script returns the actual file path — callers should use the echoed path, not assume .mp3.

Usage

# Generate audio from text (auto-deletes after delivery)
~/.openclaw/skills/audio-summary/bin/generate_audio "Your summary text here" --cleanup

# With custom output path
~/.openclaw/skills/audio-summary/bin/generate_audio "Text" --output /tmp/custom.mp3

# Use specific provider
~/.openclaw/skills/audio-summary/bin/generate_audio "Text" --provider openai --cleanup

Cleanup After Delivery

IMPORTANT: Audio files are transient. Delete them after delivery.

Pattern:

# Generate, deliver, delete
AUDIO_FILE=$(~/.openclaw/skills/audio-summary/bin/generate_audio "text" --provider openai)
# ... upload or serve the file ...
rm -f "$AUDIO_FILE"

Why: Audio files can build up quickly. A 30-second clip is ~400KB. Clean up after every use.

Incoming audio cleanup: Voice messages in ~/.openclaw/media/inbound/ should also be cleaned periodically:

# Clean up audio older than 1 day
find ~/.openclaw/media/inbound -name "*.ogg" -mtime +1 -delete

Script Output

Returns the path to the generated audio file (.wav for Gemini, .mp3 for OpenAI/ElevenLabs):

/tmp/audio-summary-1738123456.wav

Delivery

After generating audio, deliver it to your platform. Example using OpenClaw's message tool:

# Upload to chat platform
message --action send --channel <YOUR_CHANNEL> --target <TARGET_ID> \
  --media /tmp/audio-summary-1738123456.mp3 --caption "Audio summary"

Summary Style Guidelines

• Conversational — Not robotic, like talking to a friend
• Brief — 150 words max (30-60 seconds)
• Friendly — Warm tone
• Actionable — Highlight what was done and what's next

Example:

"Hey! Just wrapped up the audio transcription skill. It uses Gemini to convert voice messages to text. Tested it on a few sample recordings — works great. The multi-provider fallback is also wired up, so if Gemini is down, it'll automatically try OpenAI and then ElevenLabs. Everything's ready to go!"

Environment Variables

Files

• bin/generate_audio — Main TTS script with multi-provider support
• SKILL.md — This documentation

Troubleshooting

"No TTS provider available"

• Check that at least one API key is set
• Verify keys are set in your environment or ~/.openclaw/secrets/ directory

"Audio file too large"

• Keep summaries under 150 words
• Use --provider openai for MP3 (smaller than WAV)

"Gemini TTS failed"

• Falls back to OpenAI automatically
• Check Gemini API status at https://status.cloud.google.com
• Requires GEMINI_API_KEY with Gemini 2.5 Flash access