Normalizer Fixes WER Formatting in STT Evals
Gladia open-sourced gladia-normalization, a library normalizing transcripts to prevent WER penalties from formatting differences in STT benchmarks. It features YAML-configurable pipelines supporting English, French, German, Italian, Spanish, and Dutch. The tool standardizes outputs like 'It's $50' to 'it is 50 dollars at 3 pm'.






