Changelog

Added phoneme validation with provider parity tests and IPA assertions

Added provider diagnostics for runtime debugging and validation

Added provider parity benchmark with Phonodist classification support

Changed eSpeak integration to use optional espeakng-runtime with native auto selection and CLI compatibility

Changed fallback calls to preserve U+200D tied IPA output and typed provider errors

Changed benchmark version reporting to use runtime diagnostics and close reference providers

Documented system and bundled eSpeak runtime installation and ownership boundaries

Changed package version reporting to use installed metadata and setuptools-scm

Changed catalog handling for kokoro-v1 assets and corrected LexHint defaults and regional compatibility

Batched eSpeak fallback calls for lookup_many lexicon misses

Added optional eSpeak and Goruut providers with pronunciation controls and docs

BREAKING: Changed pronunciation APIs to structured variants and provider errors; added batching and JSON schema 2

Fixed IPA marker normalization, structured provenance, and exact/prefix lookup parity

Documented the Lexphon pronunciation metadata handoff for downstream adapters

Fixed support for lexicon implementations without a prefixes helper while preserving longest-prefix matching

Fixed version tag detection for tags with or without the v prefix

Added explicit CLI commands and structured diagnostics for unavailable pronunciation data

Added prefix matching and six additional language profiles

Documented package metadata and test coverage status in the project README

Added lexicon-first phonemization with layered lookup and normalized IPA output

Changed pronunciation asset handling to validate before atomic offline activation