← Back to Learning Hub

3.1 Text Preprocessing

Beginner · 1 article · MODULE 3: NLP

Transcription Error Rates: CER, WER, and What They Hide

Tutorial · Beginner · 17 min

The same transcription scores 5.26% CER and 20.00% WER from the same two errors, because one bad character ruins a whole word. Normalization and aggregation move the number more than model changes.

NormalizationEvaluation

Includes Colab notebook

By: Anacodic Team

Start Reading →
AnacodicAI Labs - Build production AI. Learn by doing.