werpy and werx
The Python packages are ecosystem siblings, with one being an ultra-fast tool for calculating and analyzing Word Error Rate (WER) and the other a complementary, easy-to-use package for lightning-fast WER analysis.
About werpy
analyticsinmotion/werpy
🐍📦 Ultra-fast Python package for calculating and analyzing the Word Error Rate (WER). Built for the scalable evaluation of speech and transcription accuracy.
Leverages C optimizations for fast sequence comparison and integrates Levenshtein distance algorithms for error analysis across strings, lists, and NumPy arrays. Provides customizable penalty weights for insertion, deletion, and substitution errors, plus built-in text normalization and detailed error breakdowns via dedicated summary functions. Designed for both single-pair comparisons and batch evaluation workflows in speech recognition and NLP model validation pipelines.
About werx
analyticsinmotion/werx
🐍📦 Easy-to-use Python package for lightning-fast Word Error Rate (WER) analysis
Powered by a Rust core, WERx delivers exceptional performance—processing ~907,000 utterances per second on LibriSpeech benchmarks with minimal memory overhead. Beyond standard WER metrics, it provides detailed word-level error breakdowns (insertions, deletions, substitutions) with customizable error weights, and integrates seamlessly with Pandas and Polars for downstream analysis. Designed for production use with minimal dependencies, it handles edge cases gracefully while supporting both single-sentence and corpus-level evaluations.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work