Dutch ASR Leaderboard

Community leaderboard for Dutch Automatic Speech Recognition models.

Live leaderboard: https://huggingface.co/spaces/tvosch/Dutch-ASR-Leaderboard

---

Benchmark Datasets

Dataset	Type	Description
FLEURS	Read speech	Parallel sentences translated from Wikipedia. Studio-quality.
VoxPopuli	Spontaneous	European Parliament speech. Formal, domain-adaptation test.
Multilingual LibriSpeech	Read speech	Audiobook style.

---

Quick Start

git clone https://github.com/tvosch/Dutch-ASR-Leaderboard
cd Dutch-ASR-Leaderboard
pip install -e .

python -m asr_nl.eval \
    --model "openai/whisper-large-v3" \
    --backend vllm \

Results are written to results/<model>.json. Add them to the leaderboard via PR.

---

Roadmap

• Dutch text normalization: Dutch-specific handling (https://github.com/ThomasKluiters/european-normalizer)
• Common Voice dataset — Or any other CV version
• Tier-2 datasets — Add CLARIN-NL datasets (N-Best 2008, CGN) with institutional access (https://opensource-spraakherkenning-nl.github.io/ASR_NL_results/)

---

License

Code: Apache-2.0. Datasets vary: see individual dataset cards.