jeanbayiha24/prototype_asr_wolof_french

Prototype d'ASR wolof : Wolof ASR + Translation Demo # prototype_asr_wolof_french : Wolof ASR + Translation Demo A simple **Streamlit web application** that allows you to: - Record or upload audio in **Wolof** language - Transcribe it using two different ASR models - Automatically translate the transcription from **Wolof → French** - Keep history of the last transcriptions & translations ## Available Models ### Speech Recognition (ASR) | Model on HuggingFace | Description | Type | |-----------------------------------------------|--------------------------------------------------|------------| | `speechbrain/asr-wav2vec2-dvoice-wolof` | Wav2Vec2 + CTC fine-tuned on DVoice corpus | CTC | | `dofbi/wolof-asr` | Whisper-small strongly fine-tuned on Wolof | Transformer| ### Translation (Wolof ↔ French) | Model on HuggingFace | Description | Base Model | BLEU (approx.)* | Direction | |-----------------------------------------------|-----------------------------------------------------------------------------|-------------------------------------|-----------------|----------------------------| | `galsenai/wolofToFrenchTranslator_nllb` | Fine-tuned NLLB for high-quality Wolof–French translation | facebook/nllb-200-distilled-600M | ~13 | Wolof → French (main) + French → Wolof | > The translation model was fine-tuned by **GalsenAI** using manually aligned parallel corpora, Common Voice, Wikipedia, administrative documents, and data collected via LinguaSprint Africa. ## Features - Microphone recording directly in browser - Audio file upload (.wav, .mp3, .ogg, .m4a) - Automatic audio resampling (→16kHz) & mono conversion when needed - GPU acceleration (`cuda`) when available - Simple translation interface (currently Wolof → French by d …