hyeonsangjeon/computing-Korean-STT-error-rates

Implements Levenshtein distance-based edit distance computation using dynamic programming to calculate character-level (CER) and word-level (WER) error rates for Korean STT outputs against reference transcripts. Handles Korean-specific preprocessing including whitespace normalization and optional punctuation filtering, plus a regex-based keyword pattern matcher that robustly detects named entities despite morphological variations (particles, endings) and spacing inconsistencies common in speech recognition results.

No commits in the last 6 months.