Hi, I did a quick test with NEUTRINO just to see how it worked. I thought it might be informative if I posted it here.
The website for neutrino can be found here: https://n3utrino.work/
It's mostly all in Japanese.
The input for the program is a .xml file exported from a music score creation program (they recommend using musescore- the file extension for that is musicxml). You can import midi into musescore, but if you have two notes next to each other at the same pitch it will combine the note into one (which is unfortunate). I entered the lyrics individually for each note in hiragana, and their example scores also use hiragana, but it looks like they have some conversions for romanji into japanese characters (I didn't look at them in detail).
To use it, put the exported score into the file named /score/musicxml which it where it defaults to find input, and change the name in the run.bat file (using notepad or something). It will be "set BASENAME=name_of_file"
You can also change the voicebank from KIRITAN to YOKO. Kiritan is like a typical pop idol sounding voice, and Yoko is like an opera singer. You can also change the pitch (default is 1) and formant (default also 1- I changed it to .9 to deepen the voice slightly).
To run, double click the "Run" file. The output will go to the "output" folder in several formats.
Anyhow, this is my test with Kiritan:
This is the raw output put over the offvocal. I didn't do any mixing or tuning. Cillia also has some tests that she tuned using vocalshifter (which sound much better).
The way I entered the lyrics probably wasn't the best. Neutrino can't seem to handle two syllables being entered for the same note (like -ai or -an) and if the notes are too short not much of a sound will be produced at all. The second syllable will be pronounced over the first one. Also Kiritan has a limited range, and if you go outside it, it will start to sound like bad shower singing (which was fully intentional by the makers of the program). Similarly if you suddenly go from a low note to a high one, also bad shower singing. Below are pictures of the score I used and the parameters for the program.
The website for neutrino can be found here: https://n3utrino.work/
It's mostly all in Japanese.
The input for the program is a .xml file exported from a music score creation program (they recommend using musescore- the file extension for that is musicxml). You can import midi into musescore, but if you have two notes next to each other at the same pitch it will combine the note into one (which is unfortunate). I entered the lyrics individually for each note in hiragana, and their example scores also use hiragana, but it looks like they have some conversions for romanji into japanese characters (I didn't look at them in detail).
To use it, put the exported score into the file named /score/musicxml which it where it defaults to find input, and change the name in the run.bat file (using notepad or something). It will be "set BASENAME=name_of_file"
You can also change the voicebank from KIRITAN to YOKO. Kiritan is like a typical pop idol sounding voice, and Yoko is like an opera singer. You can also change the pitch (default is 1) and formant (default also 1- I changed it to .9 to deepen the voice slightly).
To run, double click the "Run" file. The output will go to the "output" folder in several formats.
Anyhow, this is my test with Kiritan:
This is the raw output put over the offvocal. I didn't do any mixing or tuning. Cillia also has some tests that she tuned using vocalshifter (which sound much better).
The way I entered the lyrics probably wasn't the best. Neutrino can't seem to handle two syllables being entered for the same note (like -ai or -an) and if the notes are too short not much of a sound will be produced at all. The second syllable will be pronounced over the first one. Also Kiritan has a limited range, and if you go outside it, it will start to sound like bad shower singing (which was fully intentional by the makers of the program). Similarly if you suddenly go from a low note to a high one, also bad shower singing. Below are pictures of the score I used and the parameters for the program.
