pl-asr-bigos-tools

This repository contains tools for benchmarking ASR systems using BIGOS corpora.
BIGOS (Benchmark Intended Grouping of Open Speech) corpora collects and unifies publicly available ASR speech datasets.
Currently Polish language is supported.
BIGOS family corpora are available at the Hugging Face platform:

• BIGOS V2.
  
• PELCRA for BIGOS
  

Both BIGOS V2 and PELCRA for BIGOS corpora are intended for evaluation of community-provided ASR systems as part of the 2024 PolEval challenge.
Evaluation results on BIGOS V1 are available in the paper
Hugging Face leaderboard for systematic evaluation of publicly available ASR systems for Polish is under construction.

How to use this repo

BIGOS datasets inspection

Replicating BIGOS benchmark results for publicly available ASR systems for Polish

Adding new ASR system to the BIGOS benchmark

Adding new dataset to the BIGOS benchmark