Skip to content

SETUP-VOSK.md

Latest commit

 

History

History
147 lines (103 loc) · 3.92 KB

SETUP-VOSK.md

File metadata and controls

147 lines (103 loc) · 3.92 KB

Vosk Setup Guide

This guide provides detailed instructions for setting up Vosk speech recognition in Open-Cluely.

Prerequisites

  • Python 3.8 or newer
  • pip (Python package installer)
  • A working microphone
  • At least 2GB of free disk space (for the model)

Installation Steps

1. Install Python Dependencies

# Install Vosk
pip install vosk

# Install sounddevice for audio capture
pip install sounddevice

# Install requests (used for model download)
pip install requests

2. Model Download

When you first run Open-Cluely, it will automatically download the Vosk model (approximately 1.8GB). However, if you want to download it manually:

  1. Create a models directory in your application folder
  2. Download the model from Vosk's model repository
  3. Extract the model files into the models directory

The default model used is vosk-model-small-en-us-0.15, which provides a good balance between accuracy and performance.

Troubleshooting

Common Issues

1. No Microphone Access

Symptoms:

  • No transcription appears
  • Error messages about audio device

Solutions:

  • Check if your microphone is properly connected
  • Verify microphone permissions in your OS settings
  • Try selecting a different audio input device

2. Model Download Fails

Symptoms:

  • Error during first launch
  • Missing model files

Solutions:

  • Check your internet connection
  • Manually download the model (see Manual Model Installation below)
  • Verify you have enough disk space

3. Performance Issues

Symptoms:

  • Delayed transcription
  • High CPU usage

Solutions:

  • Close other CPU-intensive applications
  • Consider using a smaller model
  • Ensure your Python installation matches your system architecture (32/64 bit)

Manual Model Installation

If the automatic model download fails, you can install it manually:

  1. Create a directory: models
  2. Download the model from: https://alphacephei.com/vosk/models
  3. Select vosk-model-small-en-us-0.15 (recommended)
  4. Extract the downloaded archive into the models directory
  5. Verify the path structure matches: models/vosk-model-small-en-us-0.15/

Advanced Configuration

Selecting Different Models

Vosk offers various models with different sizes and languages. You can change the model by:

  1. Downloading a different model from Vosk Models
  2. Extracting it to the models directory
  3. Updating the model path in the application settings

Available model types:

  • Small models (~50MB) - Fast but less accurate
  • Medium models (~1.8GB) - Good balance (recommended)
  • Large models (~4GB) - Most accurate but slower

Audio Device Selection

By default, the system's default microphone is used. To use a different audio device:

  1. List available devices:
import sounddevice as sd
print(sd.query_devices())
  1. Note the device index you want to use
  2. Update the device settings in your configuration

System-Specific Notes

Windows

  • Ensure Microsoft Visual C++ Redistributable is installed
  • Use Python 3.8+ 64-bit version
  • Check Windows Security for microphone permissions

macOS

  • Grant microphone permissions in System Preferences
  • Install Python through Homebrew for best compatibility

Linux

  • Install PortAudio development package: 
    # Ubuntu/Debian
sudo apt-get install portaudio19-dev

# Fedora
sudo dnf install portaudio-devel
  • Ensure your user has audio device permissions

Additional Resources

Support

If you encounter any issues not covered in this guide:

  1. Check the GitHub Issues
  2. Create a new issue with:
    • Your system information
    • Error messages
    • Steps to reproduce the problem