Changelog

Updated minimum TTS generation stack to PyKokoro 0.8.3 and added kokorog2p 0.8.0 as a direct dependency

Bumped SSMD minimum to 0.8.1 and advanced renderer contract to schema 3 with kokorog2p text preparation version

Documented automatic written-to-spoken preparation, explicit SSMD say-as ownership, and minimum dependency contract testing

Changed the minimum TTS generation stack to PyKokoro 0.8.3 and kokorog2p 0.8.0 with aligned SSMD constraints

Fixed resumable conversion from mixing audio generated under incompatible written-to-spoken renderer contracts

Documented automatic written-to-spoken preparation, explicit SSMD say-as ownership, and minimum-version CI coverage

Fixed safe paragraph conversion resume from saved generation settings with field-level mismatch diagnostics

Recovery after a paragraph output directory is created before the first unit is finalized by establishing ownership metadata immediately

Resume orchestration so incompatible state reports an actionable reason instead of silently starting a fresh conversion

Paragraph-mode resume across processes by persisting an effective preparation seed per chapter before randomized short-sentence processing

Paragraph unit progress, automatic and explicit seeds, –fresh restart behavior, schema 5 limitations, and an interruption-based resume example

Added configurable SSMD prosody method selection with CLI overrides and persistent configuration

Added EPUB chapter structure and emphasis preservation through epub2text Markdown extraction

Changed TTSForge to delegate silence generation and SSMD downmixing to AudioSig

Changed Markdown EPUB extraction and emphasis preservation to be the default while keeping audible emphasis plain

Changed resume handling to invalidate flattened EPUB artifacts when the rendering source representation changes

Documented prosody method selection, EPUB emphasis detection, and SSMD emphasis policy settings

Documented the EPUB extraction, SSMD markup, and audible rendering layers

Added AudioSig integration for audio processing

Added configurable SSMD prosody method selection with CLI overrides and persistent configuration

Added prosody configuration settings from AudioSig

Added EPUB chapter structure and emphasis preservation through epub2text Markdown extraction

Added terminal width adaptation for CLI output

Added exact-line subchapter markers that become paragraph pauses during text extraction

Added configurable phrase-level short-sentence handling with advanced settings and synthesis statistics

Added chapter exclusions for converting selected ranges without specified chapters

Added selectable ONNX Runtime providers and provider aliases for command and persistent configuration

Added SSMD 0.8 validation, diagnostics, audio references, prosody controls, and marker sidecars

Added resumable book conversion that restores compatible progress, chapter selection, and output paths

Changed the CLI to typed commands while keeping help and version startup independent of backend loading

Added nested short-sentence configuration commands with validation for conflicting options

Changed SSMD emphasis to plain by default while retaining explicit approximate, warning, and rejection policies

Fixed short-sentence settings so they reach both text and phoneme synthesis

Improved WAV chapter merging without requiring an external encoder and tolerated timestamped model-cache filenames

Improved chapter audio merging with input validation, safe metadata handling, and atomic output replacement

Fixed persisted conversion settings validation and pause-related configuration handling

Improved resume portability across POSIX and Windows paths and serialized concurrent atomic replacements

Documented the Markdown migration, expanded installation and usage guidance, and added documentation validation

Fixed German voice support and added short sentence handling

Fixed Adapted pause defaults for better audio output

Fixed SSML and dictionary syntax for correct processing

Added Initial TTS-Forge release with CLI, conversion pipeline, and basic text-to-speech

Added Resume option for interrupted book conversions

Added Persistent conversion settings for resume functionality

Added Sample command for quick testing

Added ONNX runtime backend for text-to-speech inference

Added Split mode for phoneme processing

Added Phoneme statistics feature

Added Phoneme conversion and phoneme-based audio generation

Added Segment pause control for audio output

Added Native ONNX runtime integration

Added Support for Kokoro-82M-v1.0-ONNX models

Added Reader mode without audio creation

Added Language option for specifying TTS language

Added Chapter announcement feature

Added Mixed language support for multilingual text

Added Custom phoneme dictionary and automatic name extraction

Added Chunking for name extraction

Added Voice blending capability

Added Input reader supporting TXT and EPUB formats

Added SSML/SSMD document generation support

Added Generate SSMD-only mode

Changed Migrated to pykokoro as core TTS engine

Changed Adapted to new pykokoro version

Changed CLI refactored into modular command structure

Changed Pipeline rewrite for pykokoro compatibility

Changed Code refactoring

Changed Refactored pause mode and improved audio merge

Changed General refactoring and bug fixes

Changed Removed static-ffmpeg as required dependency

Changed Made sounddevice an optional dependency

Fixed Audio generation from phonemes

Fixed Duplicate chapter reading

Fixed Chapter announcement behavior

Fixed Various bugs in conversion pipeline

Fixed Generation config

Fixed Input reader issues

Fixed Title stripping in SSMD

Fixed Phoneme creation

Fixed Removed compiler warnings

Fixed Soundfile to only handle WAV files

Documented Updates and voice mixing examples

Documented Updates for voice blending

Documented Updated API documentation

Documented Updated documentation and fixed Windows unit tests