Audio File Transcriber GUI

Offline audio-to-text transcription tool powered by faster-whisper.
This desktop application transcribes local audio/video files (MP3, WAV, MP4, etc.) into English text using a local Whisper model—no internet required after the first launch.

Note: Designed for batch processing of pre-recorded content. For live streams, use the companion app Stream Recorder and Transcriber.

---

✨ Features

• Supports multiple formats: MP3, WAV, MP4 (any file with audio track)
• Automatic audio conversion: resamples to 16 kHz mono using PyAV
• Accurate transcription using Whisper small model (int8, CPU-friendly)
• Sentence-by-sentence output with visual formatting (red → blue)
• Export to .txt with same name as input file (e.g., talk.mp4 → talk_tr.txt)
• Language mode selection:
  • English → English (forced)
  • Any language → English (auto-detect + translate)
  • Russian → English (explicit)
• Real-time timer and progress feedback
• Graceful stop with logging of partial results
• Temporary file cleanup after processing

---

🧠 Model Details

• Uses faster-whisper with the small model (better accuracy than base)
• CPU-only, int8 quantization for low memory usage (~1–2 GB RAM)
• First run downloads ~500 MB model to ~/.cache/huggingface/hub
• Model is loaded once per session

---

📦 Requirements

• Python 3.8+
• Required packages:

  pip install faster-whisper pyside6 pyaudio pydub numpy av

• ~600 MB free disk space (for model + temp files)
• No FFmpeg required — audio conversion handled by PyAV

---

🚀 Quick Start

1. Download or clone the project.
2. Install dependencies:

   pip install -r requirements.txt  # if created

3. Run the app:

   python file.py

4. Click File, choose an audio/video file.
5. Select language mode and click START.

On first launch, the small Whisper model will be downloaded automatically.

---

📁 Output

• Text: displayed in real time; each complete sentence appears in red bold, then turns blue bold.
• Transcript file: saved as {original_name}_tr.txt in the same directory as the executable.
• Log: file.log with entries like:
  2025-12-11 15:22:10, meeting.mp4, Length: 42:18, Duration: 03:25, Transcription completed

Log file uses rotating handler (max 5 MB, 3 backups).

---

🛠️ UI Controls

Element	Function
Language dropdown	Choose transcription/translation mode
File browser	Select MP3/WAV/MP4 (or any AV file with audio)
START / STOP	Begin or interrupt processing
Timer	Shows elapsed processing time
Text box	Live transcription output

---

⚠️ Notes

• Do not close the app abruptly during processing—use STOP to ensure partial results are saved.
• The app splits text at sentence boundaries (. ! ?) for clean output.
• Input files with no audio track will trigger an error.
• For best results, use clear, speech-focused recordings (e.g., interviews, lectures, podcasts).

---

📜 License

This tool is for personal or research use.
Underlying libraries:

• faster-whisper — MIT License
• PyAV — BSD License
• PySide6 — [LGPL/GPL]

---

Built with Python, faster-whisper, and PyAV.
Version: V201125

---