TGSpeechBox v3.0-beta3 is out! 19 bug fixes, 8 new features, 5 language pack improvements.
The big ones: stop consonants now use research-based burst spectral templates from Stevens & Blumstein — alveolar, velar, and labial stops each get their own shape, so /d/ vs /g/ and /t/ vs /k/ are clearly distinct. Stop clusters in words like "locked" and "kept" now properly unrelease the first stop, the way natural speech does.
MOUTH diphthong onset was only 30 Hz from schwa — "outside" sounded like "ertside." Fixed with Hillenbrand GenAm data. Per-diphthong duration scaling replaces the old global knob, so PRICE gets the time it needs without bloating GOAT. Diphthong rate compensation keeps bare "I" and "Y" from losing identity at high speech rates.
New Fujisaki clause-type overrides let language pack authors tune question/exclamation intonation in YAML. Spanish gets proper Castilian vs Latin American approximant splits. Australian English recovers its hand-tuned vowels.
Clause-final sonorants no longer clip. Cascade resonator pops, gone. Tap timing, fixed three ways.
And yes — we know en-gb PRICE still sounds a bit Stewie Griffin. The glide doesn't curve down the way it should yet. We hear you, it's on the workbench.
https://github.com/tgeczy/TGSpeechBox/releases/download/v-300b3/TGSpeechBox-v300b3.nvda-addon
https://github.com/tgeczy/TGSpeechBox/releases/download/v-300b3/TGSBPhonemeEditor-v300b3.zip
https://github.com/tgeczy/TGSpeechBox/releases/download/v-300b3/TGSpeechSapiSetup-v300b3.exe
https://github.com/tgeczy/TGSpeechBox/releases/download/v-300b3/TGSpeechBox-v300b4.apk
https://github.com/tgeczy/TGSpeechBox/releases/download/v-300b3/tgspeechbox-linux-aarch64-v-300b3.tar.gz
https://github.com/tgeczy/TGSpeechBox/releases/download/v-300b3/tgspeechbox-linux-x86_64-v-300b3.tar.gz
https://testflight.apple.com/join/jvvGY6Fz
@Tamasg This is starting to sound really good, might throw this on my apple devices and try to get used to it. Also the Polish support is starting to sound better, but there are many phonemes that are wrong. Things like the R should be rolled, which I can hear the synth can very much do because it sounds great in Spanish. Also, the letter Y. It should nearly always be pronounced with an "ih" sound like in the word whip, where right now it sounds more like an oo, you can test this with a word like "my." Also, ę is literarly just pronounced as n where the sounds are definitely different but I have no idea how to explain that in words lol. I might try to contribute fixes if when I have more time to figure out the phoneme editor. But you are doing some really amazing work here.
@pitermach oh thank you. Your words already said a million things there that I can tune, so huge huge thanks. My biggest Polish contributor (who I've ignored, but won't for the next beta) has been @spacepup - he's tuned a lot of it and helped get it in better shape. The Y sound is something he has in his new pack actually fixed, and the rolling of the R thing, I can for sure work on as you said. So yeah, expect Polish pack updates now, especially as I've gotten feedback from two solid native speakers. I'm always afraid to tune these packs on my own because I'm only Hungarian, and so I can tune my own language well enough, but I leave Spanish and Portuguese up to the people who actually know them. Lol like @clv1 - that's why I rarely touch them. But yeah, again, huge huge thanks for this.