Module postprocess 

Shared transcript cleanup applied after every provider (JOE-1609).

Whisper (and some remote models) emit control / non-speech tokens that should not leak into scripted txt/srt/json output. Normalization produces an explicit NormalizationReport listing repairs and drops.

Structs

NormalizationEvent

One normalization repair or warning (deterministic, no provider payloads).

NormalizationReport

Report of repairs applied while normalizing a provider result.

Functions

detect_repetition_degeneration

Conservative transcript degeneration detector (JOE-1650).

normalize_result

Clean provider output into a stable, scriptable result.

normalize_result_with_report

Like normalize_result but returns an explicit repair report.

truncate_chars

UTF-8–safe truncation for error messages (never panics on char boundaries).

validated_segment

Validated segment constructor (rejects NaN/inverted spans).