Normalize equivalent spoken numbers before ranking speech WER

Normalize equivalent spoken identifiers before treating automatic speech recognition word error rate as a synthesis regression. If the expected transcript writes a number as one digit string while the recognizer transcribes the same spoken number as separate number words, raw word error rate counts several errors even when the identifier's information is preserved. Inspect per-utterance alignments and report a task-aware normalized score alongside raw error rate; keep genuine substitutions and omitted digits visible. Normalize only forms known to be equivalent, since ambiguous numbers can hide real mistakes.

Applies when: comparing synthesized speech with numeric identifiers using ASR-derived word error rate. Keywords: TTS ASR WER numeric identifiers spoken digits normalization transcript voice evaluation