VocaVerse Network

in Vocaloid engine noise is when you hear a background noise, people that have managed to work with V3devkit say that THAT happens when you do a bad config, is over stretching so the engine noise took presence

in voiceroid, is Voiceroid 1, ex sometimes the voice in some words sound like is drowning, the perfect example of this is spaming lines over a single vowel, since the vowel repeats you can hear a metalic noise of the program strugling to recreate that sound [あーーーーーーーーーーー]

To me it's kind of like a little quirk in the vocals that tends to come from the particular method of synthesis, but I'm not good at putting it into words. For example, Vocaloid specifically seems to add a kind of... edge, I guess, to the vocal. I remember one time I was talking on this forum about someone on Twitch recognizing an Oliver song as the same voice as Puppycat, and someone commented that when they watched Bee and Puppycat that they recognized Puppycat was a Vocaloid from the "weird envelope sound." If I had to guess I think it comes from the way the program splits up phonemes? But I'm definitely not technologically savvy enough to be sure or explain it well. It also seems to make vocals sound more nasally? Like it messes with the timbre of the voice provider's voice in the recordings in a specific way.

CeVIO, especially pre-AI, has like a... buzzy/shaky sound? UTAU, I'm not sure if what I think of when I think "UTAU engine noise" is really engine noise or noise from typically-not-professional-quality recordings. Honestly if I try to listen really hard I can kiiind of maaaybe hear a noise from Synth V, but it's so faint it's even harder to try to put into words, other than "It's simply not a real human, it's not going to sound perfectly like one."

All of that said, I don't think it's a big deal usually, I just felt like trying to put what "engine noise" means/sounds like to me in words X'P

i once saw someone say utau engine noise sounds like running water, and i think that's an apt comparison (ofc keeping in mind that diff resamplers sound different)
v6 in particular (not vx beta) almost sounds like it's autotuned in places?

CeVIO CS7 definitely has a slight buzz to it and V4 has a slight roughness (idk if that’s the right word) that V5 doesn’t have. Plus V5 handles pitch bends and note bends differently so the rendered result feels slightly different than V4. Engine noise doesn’t generally bother me except for Utau sometimes. Although if it’s really a problem you can always swap the resampler.

It really depends on the synth. In Vocaloid1, I hear a noticeable "violin" sounding kind of timbre to each voices, most noticeable with Meiko. Miriam has a clicky sound to her consonants, the breathiness parameter sounds like bassy static, and you can hear how the program tries to "fade" on the end of the vowels to connect to the next consonant. It makes it sound rather jumpy.
Vocaloid2 is different as you can hear the short vowel sample being looped to fit the note, it makes the vowel sound metallic, unnaturally clean and consistent in volume.
Megurine Luka has a popping sound to her consonants most noticeable when she says "t" or "k", which is very difficult to recreate by a human so consistently.
Once you start adding things such as breathiness, you can hear an emphasis in certain frequencies that sounds like a sharp static. It's more noticeable with English voices. The engine sounds like a clarinet.

Piapro NT has a cold-sounding high-frequency buzzing noise like an UTAU, and the vowels are timestretched to fit the notes, which sounds different than Vocaloid2's looping, there's also some noticeable loudness compression. The consonants are noticeably very emphasized, sounding clicky with a static behind it. The vowels have been pitch-corrected so that they sound perfecty flat and autotuned, and there's no volume fading (attack, hold or decay) in the vowels, making long notes sound very exaggerated. Miku NT may sound like a kazoo.

@Twillby I think I'm the "weird envelope sound" person lol. Synthesizers work by feeding an oscillator signal (or a sample) into an amplifier controlled by a volume envelope, which controls the attack/decay/sustain/release properties, shaping the sound. I'm not sure what combination of situations provoke it, but Vocaloid sometimes uses too soft of an attack (which is unnatural compared to a human voice), so you can sort of hear the "voice" going from silent to its peak volume during the start of the note. Like the character of a bow slowly being drawn across a string, instead of a sharp pluck. Puppycat is set up somehow to maximize that effect, or maybe Oliver just tends to do that more.

I'm not sure how I'd go about controlling that "metallic" sound Vocaloid 2-3 had more with V4, but I kind of like it and it might be fun to try and maximize it.

A lot of the noise I tend to notice (and want to remove) is fairly easily controllable, because it has the same properties of "room noise" in a vocal recording. Especially on lower notes, you sometimes get a low buzzing, which can usually be removed with EQ. I suspect much of that sort of noise is artifacts from the recording process; the sampling engine adjusting the samples probably distorts stray reverb and other things picked up by the mic