Voice To Text

Convert voice messages and audio files to text using Vosk offline speech recognition. Use when a user sends a voice message, audio file, or asks to transcribe speech to text.

Convert voice messages and audio files to text using Vosk, an offline speech recognition toolkit.

Setup

1. Install dependencies:

   # macOS
   brew install ffmpeg
   pip install vosk
   
   # Linux
   apt-get install ffmpeg
   pip install vosk
   

2. Download a Vosk model:

   mkdir -p ~/.vosk/models && cd ~/.vosk/models
   
   # Chinese (small, fast)
   curl -LO https://alphacephei.com/vosk/models/vosk-model-small-cn-0.22.zip
   unzip vosk-model-small-cn-0.22.zip
   
   # English (small)
   curl -LO https://alphacephei.com/vosk/models/vosk-model-small-en-us-0.15.zip
   unzip vosk-model-small-en-us-0.15.zip
   

Usage

When the user provides a voice message or audio file path, run the transcription:

python3 ~/skills/voice-to-text/transcribe.py "<audio_file_path>"

For specific model selection, set the environment variable:

VOSK_MODEL_PATH=~/.vosk/models/vosk-model-cn-0.22 python3 ~/skills/voice-to-text/transcribe.py "<audio_file_path>"

Supported Audio Formats

• MP3, WAV, M4A, OGG, FLAC, AAC, WEBM
• Voice messages from WeChat, Telegram, WhatsApp, etc.

Available Models

Model	Language	Size	Notes
vosk-model-small-cn-0.22	Chinese	42M	Fast, good accuracy
vosk-model-cn-0.22	Chinese	1.3G	High accuracy
vosk-model-small-en-us-0.15	English	40M	Fast, good accuracy
vosk-model-en-us-0.22	English	1.8G	High accuracy

Download models from: https://alphacephei.com/vosk/models

Example Workflow

1. User sends a voice message via WeChat/Telegram
2. OpenClaw receives the audio file
3. Run: python3 transcribe.py /path/to/voice.ogg
4. Return transcribed text to user

Troubleshooting

• No model found: Download a model to ~/.vosk/models/
• ffmpeg not found: Install via brew install ffmpeg or apt install ffmpeg
• Poor accuracy: Try a larger model for better results

Notes

• Works completely offline after model download
• Supports multiple languages (download appropriate model)
• Audio is converted to 16kHz mono WAV for processing