Hacker News new | ask | show | jobs
by __fst__ 86 days ago
Was playing around a bit and for its size it's very impressive. Just has issues pronounciating numbers. I tried to let it generate "Startup finished in 135 ms."

I didn't expect it to pronounciate 'ms' correctly, but the number sounded just like noise. Eventually I got an acceptable result for the string "Startup finished in one hundred and thirty five seconds.
3 comments
rohan_joshi 86 days ago
yeah we're fixing this at the model level too. but in the meantime, there is a way to add text preprocessing for you, and if you have a special use-cased, claude code should be able to one-shot custom preprocessing. its the way that most existing tts models (including sota cloud ones) deal w numbers and units, they just convert it into string.
link
rohan_joshi 86 days ago
thanks a lot for trying it and giving feedback. custom preprocessing will fix this for 95% of use-cases. and as i mentioned, this will be fixed at the model level in the next release.
link
magicalhippo 86 days ago
I tried it with some "hard mode" text:

The above SECDED check-bit encoding can be implemented in a similar way, but since it uses only three-bit patterns, mapping syndromes to correction masks can be done with three-input AND gates.

It sounded quite good indeed for the normal English stuff, but I guess predictably was quite bad at the domain-specific words. It misspoke "SECDED", had wrong emphasis on "syndromes", and pronounced "AND gates" like "and gates".

Could you give some example of what kind of preprocessing would help in this case? I tried some local LLMs, but they didn't do a good job (maybe my prompts sucked).

link
nozzlegear 86 days ago
> pronounciating

I'm not sure if you're misspelling it deliberately or not, but the word you're looking for is "pronounce" and it's verb form "pronouncing", as in "It just has issues pronouncing numbers" and "I didn't expect it to pronounce 'ms' correctly."

link
fc417fc802 86 days ago
He mixed pronounce with enunciate. It's an understandable mistake IMO. (English also has annunciate. Truly a cursed language in many respects.)

https://en.wiktionary.org/wiki/enunciate#English

link