tgeczy/TGSpeechBox v-300rc2 on GitHub

TGSpeechBox v3.0-rc2

Release Candidate 2. 34 commits since RC1. Big round of DSP fixes, dictionary tooling improvements, and platform polish. Getting close.

Sound Quality

High-frequency rolloff at lower sample rates: At 22050 Hz and 16000 Hz, the DSP was too aggressive with high-frequency filtering — speech sounded muffled and lacked brightness. Reworked the sample-rate-dependent filtering so highs come through naturally at all supported rates. Thank you @Simon818 for helping diagnose this by ear.
Improved /p/ stop burst: Longer, poppier, less aspirated. "Creepy" vs "creaky" now clearly distinct instead of blending together.
Comma-list inflection fix: The first word of comma-separated lists was getting period-style falling inflection instead of a continuation rise. Disabled singleWordClauseTypeOverride which was misclassifying short pre-comma clauses.
Special coarticulation leaking across word boundaries: Formant sweeps from coarticulation rules were carrying across unrelated words, causing audible artifacts. Fixed boundary check.
FORCE vowel + rhotic tuning: Multiple passes on the FORCE vowel (/ɔɹ/) allophone rules — less pirate-sounding, smoother F3 transition, fixed rule matching for GOAT-origin words like "force" itself.
American /ɹ/ rhotic tuning: Less retroflex, smoother F3 transition into following vowels.

"Pronunciation (user)" category: New category in the dictionary editor on Android and iOS. View and manage only your own entries separately from the bundled dictionary.
Export user dictionary as TSV: Export your personal entries as full 5-column TSV, compatible with NVDA import.
25 community dictionary entries: Merged contributions from fastfinge/tgict — Harry Potter names, anime loanwords, tech terms. 46 entries and growing. Huge thanks to @fastfinge for collecting these.
Preview button in insert-phoneme picker (Win32): Hear a phoneme before inserting it.
Dictionary replacement for hyphenated words: Fixed lookup that was skipping hyphenated entries.

"Save Voice Profile Sliders to YAML As..." button: Export your current voice profile slider values as a YAML file from the settings panel.
"Import Voice Profile Sliders..." button: Load a previously exported voice profile YAML.
"Export User Dictionary..." button: Export your user dictionary entries as TSV.
"Import User Dictionary..." button: Import a TSV dictionary file into your user dictionary.

Android: Added x86_64 architecture support for Chromebook users. versionCode 334, versionName 3.0-rc2.1, assetVersion 18.
Linux: Installer now auto-configures Speech Dispatcher integration — detects speechd.conf, installs the generic module config, and optionally sets TGSpeechBox as default synthesizer. Duplicate-safe on reinstall.
Hungarian: New e_hu phoneme for /eː/ (é vowel) with language-specific formants.

IPA injection for parenthesized words (issue #69): Words in parentheses now correctly receive IPA overrides on NVDA and SAPI.
Inflection 0% not producing true monotone (issue #68): The inflection scale slider at 0% still had residual pitch movement. Now fully flat.
thought_pre_rhotic_close allophone rule: Was dead config since it was added — rule conditions never matched. Fixed.
tgsbRender --prepare-text: Fixed argument count mismatch in setLanguage call.