Whisper Hebrish: Side By Side Demo
Compare the performance of Whisper-Hebrish (fine-tuned for mixed language speech) with the stock Whisper Large V3 Turbo model.
Record or upload audio to see side-by-side transcription results.
✨ WHISPER-HEBRISH (Fine-Tuned)
📦 WHISPER LARGE V3 TURBO (Stock)
About This Demo
This Space provides a side-by-side comparison of automatic speech recognition between:
- Whisper-Hebrish: A fine-tuned version optimized for mixed Hebrew-English speech patterns
- Whisper Large V3 Turbo: OpenAI's stock model for general-purpose transcription
How to Use
- Record audio using your microphone or upload an audio file
- Click "Transcribe with Both Models"
- Compare the results side-by-side
About Whisper-Hebrish
Whisper-Hebrish is a fine-tuned version of OpenAI's Whisper model, specifically optimized for handling mixed language speech patterns common in Hebrew-English conversations and switching between languages.
Model Information
🤗 Model: danielrosehill/Whisper-Hebrish
The fine-tuned model has been trained to better handle:
- Mixed Hebrew-English speech
- Switching between Hebrew and English
- Hebrew-specific phonetic patterns
- Common transliteration patterns
Technical Details
- Inference powered by Zero GPU
- Both models run simultaneously for direct comparison
- Supports both microphone recording and file upload
- Based on Whisper Large V3 architecture
Use Cases
This comparison tool is useful for:
- Evaluating model performance on bilingual content
- Testing speech recognition accuracy for mixed-language scenarios
- Understanding the benefits of fine-tuning for specific language patterns
- Benchmarking ASR quality for mixed Hebrew-English speech
Compare accuracy, handling of language mixing, and transcription quality between the fine-tuned and stock models.