jonatasgrosman/wav2vec2-large-xlsr-53-dutch
🧠 AI Modeljonatasgrosman
Fine-tuned XLSR-53 Dutch speech recognition model, state-of-the-art on Common Voice 6.0.
The model is based on the wav2vec2-large-xlsr-53 architecture, which uses self-supervised learning from raw audio. It was fine-tuned using CTC (Connectionist Temporal Classification) loss on the Dutch language data from Common Voice 6.0. The model supports end-to-end speech recognition and can be used via the HuggingFace pipeline for inference. It has been evaluated on the Robust Speech Event and is part of the HF ASR Leaderboard. With over 2.6 million downloads, it is one of the most popular Dutch ASR models available open-source.
💡Highlights
- ├─Fine-tuned on Dutch Common Voice 6.0
- ├─WER ~8% on test set
- └─2.6M+ downloads, open-source
🎯For
- ├─Dutch NLP Researchers
- ├─ASR Developers
- └─Dutch Language Tech Enthusiasts