search

mathigatti / RealTimeSingingSynthesizer

Live Coding Singing Synthesizer. Python sinsy-NG wrapper.
GNU General Public License v3.0
59 stars 6 forks source link

MBROLA voices - no sound #5

Open kormoczi opened 3 years ago

kormoczi commented 3 years ago

Hi,

Is there a way to use MBROLA voices? I have tried to test this in Colab (installed mbrola and the mbrola voices as well), but the result is empty (the length of the output looks fine, but there is just silence).

Another question: Do you have plan to make the notes of the singing connected to the syllables instead of the words?

Thanks!

mathigatti commented 3 years ago

Hi!

I also failed trying MBROLA in the past unfortunately and I'm too lazy to debug C code right now.

About the notes per syllable it should be doing that. Long words should be splitted into syllables if you have many notes to sing. You can try visualizing the musicxml file with musescore to check it. Let me know if you find some issue with that.

kormoczi commented 3 years ago

Thanks for your answer, @mathigatti, I can understand the situation with the MBROLA voices.

About the long words... Can you suggest a way, how I can split them into syllables, and still keep the original sounding of the words?

Thanks!

mathigatti commented 3 years ago

The software should do that for your automatically, let me know if there is something specific you want to do that's not working and I will give you an example of how to do it

I mean if you write "artificial" and you want it to sing 4 notes, it's going to split it into "ar" "ti" "fi" "cial". If you want it to sing 6 notes it will do something like "ar" "ti" "fi" "cia" "aa" "al"

Anyway I'm testing it and it looks like, as you say, it doesn't keep the original sounding of the words. You can check this other project which has a slower synthesis time but better singing quality