whispa-app 2.1.0

GUI for Whisper transcription & MarianMT translation

Whispa App

Audio Transcription & Translation Tool
Version: 2.1.0

---

Overview

Whispa App is a desktop GUI for:

• Transcribing audio files to text using OpenAI’s Whisper models
• Translating the transcribed text into multiple target languages via MarianMT

Built with:

• Python 3.10+
• CustomTkinter (modern Tkinter theming)
• PyTorch and faster-whisper for transcription
• Transformers and MarianMT for translation
• psutil for live system stats

---

Key Features

• Five Whisper model sizes: tiny, base, small, medium, large
• Translate into Spanish, French, German, Chinese, Japanese
• Advanced settings: VRAM threshold, beam sizes, VAD filter, temperature, length penalty
• Progress bars and real-time status updates
• Local caching for offline use after initial download

---

System Requirements

• OS: Windows 10 or later
• Python: 3.10 or higher (if installing via pip)
• CPU only by default; GPU supported via extra install
• Internet: Required only for first-run model downloads

---

Installation

📝 Option A: Windows Installer (Recommended)

1. Download WhispaApp-2.1.0-Setup.exe.
2. Run the installer and follow the prompts.
3. A console window will show pip and model download progress.
4. Launch Whispa App from the Start Menu when done.

🐍 Option B: pip (Requires Python Installed)

# CPU-only
pip install whispa_app[cpu]

# (Optional) GPU support
pip install torch --index-url https://download.pytorch.org/whl/cu118
pip install whispa_app[gpu]


# Download all models (first time only)
whispa-prefetch

# Launch the GUI
whispa


Quick Start
Browse for an audio file (.wav, .mp3, .m4a).

Select a Whisper model size and click Transcribe.

Choose a target Language and click Translate.

(Optional) Open Advanced to tweak VRAM, beam sizes, VAD, etc.

Save results via File → Save Transcript/Save Translation.

Advanced Settings
Setting   What it does
Min GPU VRAM (GB)   Minimum VRAM before falling back to CPU inference
Transcription Beam  Beam width for Whisper (higher = more accurate, slower)
VAD Filter     Skip silent segments during transcription
Translation Beam    Beam width for MarianMT translation
Length Penalty Penalizes shorter/longer translations (⧸1 favors longer output)
Temperature    Sampling “diversity” parameter for translation
Hover any control in the app for a tooltip with details.

First-Run Model Download
On first launch, Whispa App will automatically:

Download Whisper weights for all five sizes

Download MarianMT models for each supported language

Models are cached under %USERPROFILE%\.cache\huggingface and used offline thereafter. If a download fails, you’ll see an error dialog—just reconnect and retry.

Troubleshooting
“CMake” or “SentencePiece” errors when installing via pip?
Ensure you have a prebuilt wheel:

bash
Copy code
pip install sentencepiece
Or use the Windows installer to avoid build-from-source.

GPU not detected?
Install the CUDA-enabled PyTorch wheel:

bash
Copy code
pip install torch --index-url https://download.pytorch.org/whl/cu118
Still stuck?
Open an issue on GitHub or email below.

Support & Contribution
GitHub: github.com/damoojeje/whispa_app

Email: [e-mail]

Contributions and feedback are welcome! Feel free to submit issues or PRs.