So I guess there are no people covering this, and yeah, I made this thread.
There's a new, deep-learning-based voice synthesizer called Neutrino. Developed by Shachi, and released freely to the public since yesterday.
It's kinda different to use, so you must prepare a MusicXML file to be synthesized, set some stuff, and it takes time to render the voice. And yeah, the performance depends on your PC specs, so you'll need a PC with 8 gig or more memory, i5/Ryzen 5 or better, and at least 10 gig storage to get somewhat desirable results, or else it takes forever to render.
Thanks to the deep-learning technique, it's kinda successful to capture the natural singing style from the original voices (though some people force-tuned it with 3rd party programs to make it somewhat smooth or customize the singing style). The results kinda reminds me with that Hoya's VoiceText where from the simple MusicXML with no pitchbend parameter, the program add the pitchbend itself and yeah.And people who are too lazy to tune can take benefits with this program. And though this program still have that noise quirk, idk if it's just me but this is my impression on any deep-learning-based voice synthesizers.
It comes with 2 VBs. The one is Youko (it's the same with the Sinsy VB)
And, you know this character, it's Tohoku Kiritan! (it's based on that publicly-released singing dataset)
You could try this by yourself by visiting the official site and download the program.
There's a new, deep-learning-based voice synthesizer called Neutrino. Developed by Shachi, and released freely to the public since yesterday.
It's kinda different to use, so you must prepare a MusicXML file to be synthesized, set some stuff, and it takes time to render the voice. And yeah, the performance depends on your PC specs, so you'll need a PC with 8 gig or more memory, i5/Ryzen 5 or better, and at least 10 gig storage to get somewhat desirable results, or else it takes forever to render.
Thanks to the deep-learning technique, it's kinda successful to capture the natural singing style from the original voices (though some people force-tuned it with 3rd party programs to make it somewhat smooth or customize the singing style). The results kinda reminds me with that Hoya's VoiceText where from the simple MusicXML with no pitchbend parameter, the program add the pitchbend itself and yeah.
It comes with 2 VBs. The one is Youko (it's the same with the Sinsy VB)
And, you know this character, it's Tohoku Kiritan! (it's based on that publicly-released singing dataset)
You could try this by yourself by visiting the official site and download the program.
Last edited: