Whisper Interface v2

A GUI application for transcribing audio/video files using the specialized Whisper Large V3 Turbo model.

Features

• User-friendly GUI: Simple and intuitive interface
• GPU acceleration: Faster performance with GPU, fallback to CPU if unavailable
• Wide format support: Works with multiple audio and video formats
• Automatic file splitting: Handles large files seamlessly
• Multiple output formats: Export transcriptions in XLSX, SRT, TXT, VTT, TSV, or JSON
• Timestamps toggle: Option to include or exclude timestamps in exports (enabled by default)
• Dark/Light mode: Choose your preferred theme
• Progress tracking: Real-time progress updates with sound notifications
• 100% offline processing: No internet required after model download
• Color-coded progress: Easy-to-read progress updates in the terminal
• Detailed logging: Logs saved locally in the /logs directory

Installation & Usage

Windows

1. Install Python 3.8 or newer from python.org.
2. Install ffmpeg by running the following commands in Command Prompt (as Administrator):

   winget install ffmpeg
   setx PATH "%PATH%;C:\Program Files\ffmpeg\bin"

3. Double-click install\setup.bat to:
   • Create a virtual environment
   • Install all required dependencies
4. Double-click install\setup_download_model.bat to:
   • Download the Swiss German Whisper model (approximately 3GB)
5. Double-click run.bat to start the application.

macOS/Linux

1. Install Python 3.8 or newer.
2. Install ffmpeg:
   • macOS: brew install ffmpeg
   • Linux: sudo apt install ffmpeg (or equivalent for your distribution)
3. Make the scripts executable:

   chmod +x install/setup.sh run.sh

4. Run the setup script:

   ./install/setup.sh

   This will:
   • Create a virtual environment
   • Install all required dependencies
5. Download the model:

   ./install/setup_download_model.sh

   • This will download the Swiss German Whisper model (approximately 3GB).
6. Start the application:

   ./run.sh

Changing the Model

The application uses the Swiss German Whisper Large V3 Turbo model by default. If you want to use a different Whisper model, you can modify the hardcoded model ID in the setup_download_model.bat (Windows) or setup_download_model.sh (Linux/macOS) file.

• Windows: Open install\setup_download_model.bat and change the MODEL_ID value:

  set MODEL_ID="your-new-model-id"

• Linux/macOS: Open install/setup_download_model.sh and change the MODEL_ID value:

  MODEL_ID="your-new-model-id"

You can find other Whisper models on Hugging Face.

Usage

The application works completely offline after the initial setup. No internet connection is required for transcription.

1. Launch the application by running run.bat (Windows) or ./run.sh (Linux/macOS).
2. Click "Choose audio files" to select one or more audio/video files.
3. Use the "Include Timestamps" checkbox to toggle timestamps in the exported files (enabled by default).
4. Select the desired output format(s).
5. Click "Start" to begin transcription.

The transcribed files will be saved in the same directory as the input files, with the same name but different extensions based on the selected output formats.

Configuring Audio Settings

You can configure several audio-related settings in the file src/audio_config.py. These settings are explained in detail within that file. Please refer to it for further customization options.

Supported File Formats

Audio

• MP3, M4A, M4B, M4P, FLAC
• OGG, OGA, MOGG, WAV, WMA
• MMF, AA, AAX

Video

• MP4, M4V, MKV, WEBM
• AVI, MOV, WMV, FLV
• And many more video formats

Output Formats

• XLSX: Excel spreadsheet with timestamps and text
• SRT: SubRip subtitle format
• TXT: Plain text
• VTT: WebVTT subtitle format
• TSV: Tab-separated values
• JSON: Structured data format

Changes from Original Version

• Runs the model on GPU, with fallback to CPU
• Uses the specialized Swiss German Whisper Large V3 Turbo model
• Improved code organization with separate modules
• Enhanced progress tracking
• Better error handling and resource management
• Added a GUI checkbox to toggle timestamps in exports (enabled by default)

Credits

• Model: Whisper Large V3 Turbo Swiss German
• Original Whisper model: OpenAI
• Whisper Interface v2: Dimitri Gerster (gerdix)

Known Bugs

• Button 'Abort' doesn't work
• App looses connection after first transcription is finished
• App has no clean Exit Button