User blog comment:Angel Emfrbl/CFM broken record issue +the issue with other synths/@comment-39901408-20190726091930/@comment-53539-20190726100826

My issue with UTAU is not "its sucky" but that you can't talk about it. None of the UTAU users ever want to admit its bad and worst then Vocaloid. Don't get me wrong, Vocaloids got a slight scratchiness to it, especially vocals with a airiness to them.

I had a conversation on Discord with someone (no names here) about how Prima can outsing a lot of Vocaloids. The mentality "all Vocaloids are good if you use them right" is indeed true... But because of the samples, things like singing styles matter. For Prima, IA, Sachiko and other such vocals, because of their samples are singing results which are professional, this translated into Vocaloid as well. So when used right, all of these vocals will know the pop Vocaloids (which amount to about half of Vocaloid) out of the water. At the end of the day, samples are where the software draws the voice from for each Vocaloid, otherwise they'd all be the same. Miku herself in V2, couldn't actually sing believe it or not and the best CFM V2 singer was Luka for a reason. And sadly, when Vocaloid sings well, it will knock UTAU out of the water.

Not only that but as I said on Discord also, to make a Vocal that could hope to beat a real singer... You've got to look at Elvis Project and what it achieved. Elvis was the research project that lead to Project Daisy, which in turn officially became Vocaloid. Elvis achieved the impossible by producing a decent vocal synth but its methods were too much. To make the vocal syntn you had to record every note, every tone variation, every pronication... For 1 song. It wasn't realistic at all. But it highlights the problem with all vocal synths, in that were have a general vocal. Yet for a synth to fully replace a real singer, in truth your looking at a voicebank made for each song because a real singer is capable of using so many types of tones and variations in a song its difficult to know how to hit everyone.

https://www.youtube.com/watch?v=BREsVS3kBeY

This is an example of how a singer can very their voice a lot... For a Vocaloid to do WONDER MOMO-i it would need a "soft" vocal and "power" vocal. So every single released Vocaloid is out straight away, sorry Rana, sorry Sachiko, etc. But thats the problem also, no single voicebank can do everything. But if you imagine throwing that song into Elvis... You've not only going to have to make those two vocals, but the vocalist hare has variations for high and low in her voice... So your looking at more then what you first think.

Once you understand the problem Vocaloid faces, and in turn all synths... You sort of understand what the implications are behind everything. To be honest, the singers not that good in that song, she is voice acting more then singing. You can hear some bad note every so often, which again... If the samples are bad you'll get bad notes... But with a nature singer, a performance can be a one-off and thus bad notes are to be expected and may even for the sake of the song be on purpose. Software leans towards perfection, but the human voice... Its not perfection.