DOCSEARCHbyAlgolia
Submit Docs

whisper-standalone-win

repository·main·Indexed 25 days ago

https://github.com/purfview/whisper-standalone-win

A collection of standalone Windows, Linux, and macOS executables for OpenAI's Whisper and Faster-Whisper. It provides high-performance speech-to-text transcription without requiring a Python installation, featuring versions such as Faster-Whisper-XXL with audio preprocessing, multiple VAD methods, and speaker diarization.

Tokens
628
Snippets
0
Records
5
Agent score
35%

What's inside whisper-standalone-win

  1. Overview of Whisper Standalone Executables

    main

    This project provides standalone x86-64 executables for OpenAI's Whisper and Faster-Whisper, allowing users to perform speech-to-text transcription without installing Python. These tools are designed for use in command-line interfaces or integrated into third-party applications like Subtitle Edit, Tero Subtitler, FFAStrans, AviUtl, and PotPlayer.

    Key Versions:

    • Faster-Whisper: Faster and more efficient than vanilla Whisper; requires less RAM/VRAM. (Note: This version is deprecated in favor of Faster-Whisper-XXL).
    • Faster-Whisper-XXL: The recommended version. Includes advanced features like audio preprocessing with MDX23 Kim_vocal_v2, multiple VAD (Voice Activity Detection) methods, and Speaker Diarization.
    • Whisper: The vanilla, unmodified implementation of OpenAI's Whisper.
    • Faster-Whisper-XXL Pro: A non-public version for project donators containing additional advanced features.

  2. System Compatibility and Installation Notes

    main

    Compatibility

    • Faster-Whisper: Windows 7, Linux v5.4, macOS v10.15 and above.
    • Faster-Whisper-XXL / Pro: Windows 10, Linux v5.15 and above.
    • Whisper (Vanilla): Windows 10 and above.

    Important Usage Notes

    • Installation Location: Do not copy the programs into Windows system folders. If you must run them from such locations, you may need to run as Administrator.
    • GPU Acceleration: The programs will automatically attempt to use the GPU if a CUDA-compatible environment is detected.

  3. Configure Faster-Whisper Models

    main

    Faster-Whisper executables look for models in a local directory relative to the executable. The default path structure is _models\faster-whisper-[model_name] (e.g., _models\faster-whisper-medium).

    Models can be:

    1. Downloaded automatically by the executable.
    2. Downloaded manually from Hugging Face (Systran or Purfview repositories).

  4. Run Faster-Whisper-XXL via Command Line

    main
    You can use the faster-whisper-xxl.exe executable to transcribe files or entire folders. The program automatically detects CUDA to utilize the GPU if available. For high-quality transcription, it is recommended to use a model size of medium or larger.

  5. Advanced Features in Faster-Whisper-XXL

    main

    The faster-whisper-xxl.exe version includes several advanced capabilities for improved transcription quality:

    • Audio Preprocessing: Uses the MDX23 Kim_vocal_v2 vocal extraction model to isolate vocals.
    • Alternative VAD (Voice Activity Detection) Methods: Supports silero_v3, silero_v4, silero_v5, pyannote_v3, pyannote_onnx_v3, auditok, and webrtc.
    • Speaker Diarization: Capability to identify and distinguish between different speakers.