README.md

metadata

title: Whisper German ASR
emoji: 🎙️
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.0.0
app_file: app.py
pinned: false
license: mit

🎙️ Whisper German ASR

Fine-tuned Whisper model for German Automatic Speech Recognition (ASR).

Description

This Space provides an interactive interface for transcribing German audio using a fine-tuned version of OpenAI's Whisper-small model. The model has been specifically optimized for German speech recognition.

How to Use

1. Upload Audio: Click on the audio input area to upload an audio file (WAV, MP3, FLAC, etc.)
   • OR -
2. Record Audio: Use the microphone button to record audio directly
3. Transcribe: Click the "Transcribe" button to generate the transcription
4. View Results: The transcription will appear on the right side

Model Details

• Base Model: OpenAI Whisper-small (242M parameters)
• Fine-tuned on: German MINDS14 dataset
• Language: German (de)
• Task: Transcription
• Performance: ~13% Word Error Rate (WER)

Features

• ✅ Upload audio files in various formats
• ✅ Record audio directly from microphone
• ✅ Real-time transcription
• ✅ Optimized for German language
• ✅ Support for audio up to 30 seconds

Technical Specifications

• Sample Rate: 16kHz
• Max Duration: 30 seconds
• Beam Search: 5 beams
• Device: CPU/GPU auto-detection

Tips for Best Results

• Speak clearly and at a moderate pace
• Minimize background noise
• Ensure audio is in German language
• Keep audio clips between 1-30 seconds for optimal results

Links

• GitHub Repository
• Model Card

License

MIT License

Acknowledgments

• OpenAI Whisper for the base model
• Hugging Face for Transformers library
• PolyAI for the MINDS14 dataset