v2.10.1

Fixes

• Synthesis: don't trim whitespace in sentence entry text in timeline

Full Changelog: v2.10.0...v2.10.1

v2.10.0

Fixes

• Fix issues with sentences spanning over multiple lines, causing breaking of segmentation
• Change to @echogarden/text-segmentation v0.4.1 which changes behavior of sentence segmentation to ensure it never spans over line breaks. Use segment boundaries from the segmentation library itself to prevent conflicts or overlapping
• Fix issues with empty timestamps in timelines. Fall back to last known timestamp if not available
• Fix issues with NaN values in generated subtitles

Behavioral changes

• Set whitespace option, which controls whitespace in timeline segment text, to default to preserve. Sentence text is not impacted by this option at the moment and always preserves whitespace
• Sentences that span over multiple lines, like in some plain-text styles, will now always break over the line. For now it requires that sentences never contains line-breaks, so older plain-texts formats like Project Gutenberg books will require some preprocessing to work correctly. This change was made to resolve significant issues with how sentences are segmented, but in the future likely enhanced to provide option to safely handle these kinds of texts

Full Changelog: v2.9.3...v2.10.0

v2.9.3

Fixes

• Fix issue with incorrect paragraph whitespace processing causing broken segment offsets and timestamps (reported in #109)

Full Changelog: v2.9.2...v2.9.3

v2.9.2

Fixes

• Fix incorrectly flipped Boolean flag causing broken word segmentation for Chinese and Japanese since v2.8.0 (reported in #108)
• Update @echogarden/espeak-ng-emscripten to 0.3.5, which fixes failure of eSpeak data file path resolution when local path contains spaces or other special characters that are escaped in standard URIs

Full Changelog: v2.9.1...v2.9.2

v2.9.1

Fixes

• Fix subtitle conversion error when an empty segment is encountered
• Use updated package for ffmpeg-6.1-win32-arm64 with fixed executable filename
• Fix compilation issues with TypeScript v5.9.x+
• Lexicon: fix IPA for "modulo"

Full Changelog: v2.9.0...v2.9.1

v2.9.0

Enhancements

• Add part callbacks to API recognize method (experimental, works only with whisper engine, callback behavior may change in future versions)
• whisper: optimized post-processing of Mel spectrogram to significantly reduce its overhead during audio encoding
• Significantly improved short-time FFT speed by using a synchronous generator instead of an async generator

Full Changelog: v2.8.7...v2.9.0

v2.8.7

Enhancements

• whisper: add token count and decoding time metrics, shown on trace log level

Fixes

• Revert to using setTimeout in sleep method, due to apparent high CPU usage with setImmediate

Full Changelog: v2.8.6...v2.8.7

v2.8.6

Fixes

• Fix Emoji patterns
• Various word and sentence segmentation fixes in @echogarden/text-segmentation

Internal

• Enhance sleep method to millisecond resolution

Full Changelog: v2.8.5...v2.8.6

v2.8.5

Fixes

• Change all dependency versions in package.json to use ~ instead of ^ versioning, meaning the stated version only applies to future bug-fix releases, like 1.2.1 and 1.2.2, but not minor version increments like between 1.2.0 and 1.3.0

Full Changelog: v2.8.4...v2.8.5

v2.8.4

Enhancements

• Update English lexicons

Fixes

• onnxruntime-node: roll-back from version 1.22.0 to version 1.21.1, due to Linux post-install script issues described in #103 (also now reported on the onnxruntime issue tracker)

Full Changelog: v2.8.3...v2.8.4