Quote:
Originally Posted by thiago.eec
The android app '@Voice Aloud Reader' lets you add corrections to your TTS engine: you can expand abbreviations [...] and correct mispronounced words by providing a more 'phonetic' alternative.
|
Thanks for that info.
Yeah, some readers may have a manual "pronunciation" override list too. I think that's a much better way to go compared to clogging up the code inside the books.
(I think TTS would be WAY better handled at the OS/reader/browser level instead of baked in—see that 2021 linked thread for some more specifics.)
- - -
Side Note: Even since that 2021 thread, there's been
enormous quality jumps in Text-to-Speech generation.
A few new tools came out which are
generational leaps above the stuff before.
The more data it can process, the better it can figure out context... automatically handling cases like "Madame" vs. "Mme." (Abbreviation) vs. "MME" (Acronym) better.
Speech-to-Text (SST) had huge jumps too.
Before, you needed a ton of training data just for a specific language... then you needed to repeat that for Languages B, C, and D... Each language was in its own silo.
Now, you mash them all together, and the network deduces patterns across nearby languages too.
(So while you're training better Spanish data... Portuguese and French and Italian are all going along for the accuracy ride too!)
SST is also auto-detecting Proper Nouns based on emphasis/speed of how the person is speaking + surrounding words. Even handling:
- Unknown words
- Book/Article titles
- Last names
- Company names
- Usernames
- + Strange/obscure terminology
... and spelling/capitalizing them perfectly.
- - -
Quote:
Originally Posted by thiago.eec
Over the time I've added a lot of these, and most of it is perfect now, with a few exceptions popping up once in a while, which is promptly corrected with a new entry.
|
Wow, I looked at their site and it links to ~10 year old Android videos... lol. Is there any more recent screenshots and things showing off those specific options/functionality?