BSC-LT/CATALAN-ASR-ARENA

Catalan ASR Arena

Leaderboard

This Space shows a public leaderboard of Catalan ASR models. It reads benchmark result JSONL shards, groups rows by evaluation, and ranks models by the benchmark metrics shown in the table.

Request creation

The public submit page sends a POST request to /api/submissions. The backend validates the payload, generates submitted_time, sets status to pending, and writes one JSONL request file to the Hugging Face benchmark dataset.

Request row:

{"model":"BSC-LT/whisper-bsc-large-v3-cat","base_model":"https://huggingface.co/openai/whisper-large-v3","architecture":"Whisper","model_size":"3GB","number_params":7.1,"model_type":"fine-tuned","languages":"catalan","submitted_time":"2025-09-29T15:08:30","license":"cc-by-0","sender":"AbirMessaoudi","status":"pending"}

Results refresh

The leaderboard refreshes benchmark result files from Hugging Face before serving /api/leaderboard, at most once every HF_RESULTS_REFRESH_INTERVAL_SECONDS seconds. The default is 1800 seconds.