Changelog

Added spaced ellipsis normalization and token realignment for model punctuation collapses

Changed German area, volume, hectare, and speed quantities to use the shared spokenform grammar

Changed the reviewed seven-language pipeline to use shared written-to-spoken preparation

Changed migrated-language written-to-spoken semantics to use Spokenform 0.2.6 and raised the abbr2words floor to 0.2.9

Added Spokenform diagnostics and provenance tracking for upstream warnings

Changed migrated-language processing order so Spokenform receives source syntax before model-punctuation cleanup

Changed Spokenform source replacement handling to preserve exact spans and retain provenance in token metadata

Changed dependency floors to abbr2words>=0.2.9 and spokenform>=0.2.6 with handoff regressions

Changed English contextual N.0 release-label pronunciation through the spokenform preparation layer

Changed German written-to-spoken semantic preparation to use spokenform while preserving kokorog2p overrides

Changed French written-to-spoken semantic preparation to use spokenform and retained number helpers as deprecated

Changed Spanish written-to-spoken semantic preparation to use spokenform and retained dialect phoneme behavior

Changed Italian written-to-spoken semantic preparation to use spokenform and retained Italian phoneme behavior

Changed Portuguese written-to-spoken semantic preparation to use spokenform and retained Portuguese phoneme behavior

Changed Czech written-to-spoken semantic preparation to use spokenform and retained Czech phonological rules

Changed English written-to-spoken semantic preparation to use spokenform and retained English G2P behavior

Changed abbr2words to serve as shared lexical abbreviation source for all seven languages

Fixed exact Spokenform replacement spans when sentence-final periods appear in structured replacements

Fixed English abbreviation and core pipeline alignment compatibility with current Spokenform semantics

Documented English semantic ownership and the Portuguese migration boundary

Documented that accepted Spokenform semantics are preserved literally in the English API documentation

Documented the abbr2words-to-spokenform-to-kokorog2p ownership boundary and seven-language migration scope

Added runtime and parity fixture assets to portable Codecrate packs

Added compact Spokenform handoff regression tests covering special characters and symbol preservation

Added dependency-free adapters for SSMD annotation spans and phrasplit-compatible segment rebasing

Changed automatic spaCy model discovery to be optional while keeping explicit model requests strict

Added valid German and French optional installation extras

Documented the SSMD-to-phrasplit-to-kokorog2p pipeline and shared clean-text coordinate contract

Added cross-package coordinate, normalization, and no-model integration coverage

Fixed French fallback voice resolution to prefer locale-specific fr-fr over bare fr, avoiding unavailable MBROLA voices

Corrected French gold lexicon nasal IPA mappings for dem̃ander, restaurant, and excellent word families

Fixed French fallback to reject empty, whitespace-only, or placeholder phoneme results instead of silently dropping OOV words

Added French fallback voice resolution, empty phoneme rejection, and gold lexicon regression tests

Add Spanish language grapheme-to-phoneme support

Added Italian language grapheme-to-phoneme support

Added Portuguese language grapheme-to-phoneme support

Added benchmark datasets for Korean, French, and Chinese languages

Added safe offline spaCy model loader to prevent network downloads in offline environments

Added factory language validation tests and offline optional dependency tests

Added release artifact checking script

Updated pygoruut dependency to version 0.8.0

Changed ruff to 0.15.0 and modernized linting configuration with isort sections and McCabe complexity limits

Changed GitHub Actions workflows for improved CI reliability

Changed Korean grapheme-to-phoneme processing with updated g2pk module and numeral handling

Changed German phoneme generation and rebuilt phoneme data

Fixed mbrola voice availability detection for unavailable voices

Fixed French fallback and phoneme generation for espeak backend

Fixed test suite compatibility for Termux environment

Fixed spaCy model download commands in CI workflows

Changed documentation format from RST to Markdown with MyST parser

Added espeak phonemizer base tests for improved backend coverage