#spamton voice if he were in a jpop boy band

[mycringefactory]

Playing around with something called Neutrino AI which is basically an AI powered version of Vocaloid lol. The audio has a female voice, male voice, then all together 😜

I thought it worked really well, given that I synced it with a random midi. If you want to try it out then here’s how I did it (to help me remember and save you some time).

Step 1: If you don’t have a song in musicxml format, download a midi.

Step 2: Download Musescore and edit the midi to be sing-able. This means only have one instrument playing, and make sure only one note plays at a time (i.e. no chords).

Edit -> Instruments -> Remove from score to remove instruments. You’ll manually have to remove notes if there’s chords.

Step 3: Export the midi as musicxml. You can do this multiple times for different instruments if you want to make an acapella or something.

Step 4: Make sure all notes (except rests) have a lyric. Currently, it has to be in Japanese.

The way I did it was directly edit the musicxml to have this within the < note > (without spaces) tags:

To add it to all notes except rests, replace all instances of < pitch > (without spaces) with:

You can also add the lyric in Musescore too if you prefer to do that.

Step 5: Put your musicxml in the “score/musicxml” folder of Neutrino. Rename it to “sample1.musicxml″ (because, for me at least, the Run script references a bunch of voice-related files needed to generate the voice as “sample1.″ If you happen to have your own version of these files, you can basically skip this step and edit the Run script to use your files instead.)

Step 6: Run the script to generate the voice. It may take a long time, especially the NSF step.

You may want to check on sample1_syn.wav while it is doing the NSF step to make sure it doesn’t sound broken and save yourself some time. It should be roughly the length of the song (or whatever length you made it).

Step 7: Check sample1_nsf.wav. If you like it, you’re done!

I noticed for the male voice, the voice kind of breaks if the notes are too high pitched. You can make the notes lower by doing this: Go to Musescore, highlight all notes, Transpose -> To Key -> Down. Listen to the song to make sure it sounds okay. If it does, export it as musicxml.

Note: By the way, the reason why I posted the XML as a screenshot instead of just text is because it breaks Tumblr. Sorry, I know it's annoying.