A desktop app for macOS and Windows runs speech recognition on the user’s computer. The first setup needs internet to install dependencies and download the model; later transcription runs locally. Local processing does not replace company rules for storing and protecting files.
Requirements
- macOS: Python 3.10+. Apple Silicon uses MLX Whisper; Intel Mac uses Faster Whisper on CPU.
- Windows: Python 3.11. Faster Whisper runs on CPU; this package does not use an NVIDIA GPU.
- Allow about 2–3 GB for the model. The first setup downloads packages and model files.
Install and launch
- Download and unzip the package to a folder you can write to.
- On Mac, open start_mac.command. On Windows, open start_windows.bat.
- Wait for the virtual environment, dependencies and model to finish installing. The model is cached for later runs.
Transcribe a recording
- Choose an audio or video file in the app.
- Select Russian, English or automatic language detection, then choose TXT or SRT.
- Start transcription and follow the progress and status messages.
- The result is saved beside the recording. The source is not modified; the app asks before replacing an existing output.
Privacy and limits
- The app does not upload audio to a public AI service. Internet is used during setup to download the model and dependencies.
- CPU speed depends on the computer and recording length. The package does not claim measured accuracy or processing time.
- Local processing does not control access to the computer, backups or output files. Follow your organization’s retention and access rules.
- Checked: Python syntax and ZIP integrity. Full transcription and the Windows installer were not run on target computers.