ill13 / SpeakLocal

A TTS extension for oobabooga text WebUI
MIT License
26 stars 1 forks source link

Improve the sound quality #4

Open ill13 opened 1 year ago

ill13 commented 1 year ago

It would be very good if it could improve the sound quality, because now it is still distorted here and there and does not "read" clearly.

Originally posted by @mykeehu in https://github.com/ill13/SpeakLocal/issues/3#issuecomment-1721870437

ill13 commented 1 year ago

More audio options added.

image

mykeehu commented 1 year ago

The sound is better. However, is it possible that the JSON encoding is not decoded back to the sound? Try this text: Elfutok a f\u00e1hoz, ahol megbesz\u00e9lt\u00fck, hogy tal\u00e1lkozunk, \u00e9s elkezdek sz\u00e1molni t\u00edzig. Amikor el\u00e9rem a t\u00edzes sz\u00e1mot, ki\u00e1ltom: "Miki, megtal\u00e1ltalak!"

Original text this: Elfutok a fához, ahol megbeszéltük, hogy találkozunk, és elkezdek számolni tízig. Amikor elérem a tízes számot, kiáltom: "Miki, megtaláltalak!"

It seems to me that it reads the codes, not decodes the accented letters before TTS.

ill13 commented 1 year ago

I'll look into improving pronuncation, however that may be well out of reach for the OS built-in TTS engines.

mykeehu commented 1 year ago

Okay, I found a solution with the help of a friend, and it now returns all unicode characters correctly.

Need replace line 66 to this and fine with hungarian text too:

string = re.sub(r'[^\w\s-]', '', string).title()
ill13 commented 1 year ago

Added string = re.sub(r'[^\w\s-]', '', string).title() and commented out the earlier regex bit.