Changelog

Improved CLI sentence splitting with Phrasplit while preserving punctuation and language-specific grouping

Added morphological metadata to token annotations and mapping-based annotation input

Added Termux installation guidance for system eSpeak and explicit compatibility-warning assertions in optional tests

Changed CLI clause batching to preserve multiline clause bodies and raised the espeakng-runtime minimum to 0.1.5

Fixed missing-phoneme warnings to report the external caller location

Added phoneme diagnostics

Added espeakng-runtime support

Updated changelog

Added inspect_espeak() diagnostics and native library auto-discovery

Changed eSpeak backend to delegate discovery and execution to espeakng-runtime

Changed eSpeak backend exports to lazy imports

Fixed CI configuration

Added a standalone voice-config-driven Piper frontend with text and eSpeak backends

Added opt-in G2Lex and Lexphon adapters with pronunciation overlays and raw-block precedence

Added reusable facades, caching, annotations, language routing, markers, and stress controls to the public API

Improved lexicon resolution, configuration validation, and compatibility checks across supported profiles

Improved package versioning with source control metadata and release-artifact validation

Changed runtime dependency boundaries and clarified optional semantic preparation

Fixed eSpeak CLI clause handling and benchmark integration

Documented batch processing usage

Documented prepared-text usage and explicit Spokenform composition