• We recently had issues with our email system, but hope it should be resolved now. If you're not receiving emails to finalize account creation, please use the Contact Us form. Thanks for your patience.
mobius017

mobius017

Aspiring ∞ Creator
Messages
1,874
Media
209
Resources
10
Reaction score
3,095
Medals
5
Joined
Last seen

Profile posts Latest activity Postings Media Albums Resources Awarded medals About

  • mobius017
    I didn't get a lot of sleep last night. Right after I laid down, I got an inspiration. Two hours later, I'd written most of a roughly 1-page fanfic.

    Which is to say: new mini vocal synth fanfic coming soon!
    mobius017
    Currently wearing moccasins (Native American shoes) and a Miku Expo t-shirt, driving coffee from my Miku mug, working on lofi (hip hop?) music featuring Australian samples and where a Japanese girl sings in English.

    The world's different aspects (not even really nationalities; all kinds of things) can get along this way, and I wish they would more often.
    • Like
    Reactions: Vector, TheStarPalace, Blue Of Mind and 3 others
    Vector
    Vector
    Music is great for that, honestly. I'm really interested in Eurodance/Eurobeat and House, which only existed because of that sort of global fusion. You have Japanese drum machines and synths (TR-909, Akai samplers, and Roland, Yamaha and Korg synths were ubiquitous) on one hand. Those genres all directly descend from disco, which predominantly was an American Black/Latin/LGBT thing to start with...and clubs like the Warehouse were key in disco evolving into early House after mainstream backlash. Then things evolved a lot in European clubs (particularly Italy, with Italo Disco and Eurobeat, and France with House acts like Daft Punk) and went on to be a major influence on J-Pop starting in the late 90s.

    Any time you hear a drum machine playing a typical four-on-the-floor dance beat, you're hearing something that's a part of that lineage. It's cool to think about.
    mobius017
    Thoughts on Miku singing in English:

    Intelligibility is important to me, but there's also a point where you have to respect a vocal synth for being what it is. Miku English is a vocal synth of a Japanese woman singing in English, which isn't her first language. That's not something you complain about; it's part of the color and part of the character.

    I personally also find her struggles in this area thematically apropos. All of her users are struggling with something, whether it's with doing the music thing or something else. The fact that English takes a bit more work for her fits well into that idea--it's like there's something she struggles with, too. That shared struggle has been part of her character since "Miku Miku no Shiite Ageru."

    That said, I wouldn't be opposed to an AI Miku who could do English more smoothly, as long as it sounds like her. There was nothing wrong with the direction of V4, IMO, for example.

    (Continued below.)
    • Like
    Reactions: Vector, WyndReed, MillyAqualine and 3 others
    mobius017
    mobius017
    Side thought:

    I wonder.... If Miku AI became a thing, and we're worried about it sounding like her, why not train the AI partly on her classic songs? Ask the big classic Miku producers-- Deco*27, kz, Mitchie M, etc.--if they'd be willing to have their 2000s-era songs used to help train Miku's AI. CFM's probably the only company with the necessary relationships with its fanbase to try something like this. Some of the producers might decline, but maybe a bunch of them wouldn't mind--maybe in exchange for modest one-time compensation, or a grateful thanks on the box. The songs are over 10 years old anyway, and producers' primary output is songs, not tuning (though it might be a somewhat bigger deal for Mitchie, who's known as a tuning god). (Or open things up and ask for volunteers to have their songs used.) Throw those songs from all those different classic era producers together, along with the human training material that would've been used anyway, into a blender and see what the AI comes out with. Shouldn't it be less AI-sameness and more of what Miku's iconic sound has come to be?

    In that scenario, you have Miku and her fanbase teaching the next generation of Miku. What could be more appropriate than that?
    AmazingStrange39
    AmazingStrange39
    YES

    I feel like her strong accent and mispronunciations give her a certain element of character? Like she worked hard to learn English herself to speak to and sing for her international fans. And I've pretty much always been a fan of characters having their own speaking styles and accents, and when it comes to Vocaloids I think about this a lot too (I did a ton of research on Irish accents largely to piece together Avanna's speaking style even).

    In a Talkloid set earlier, like in her early V3 days, I might have Miku be more likely to make mistakes; not full on "you-no-take-candle" level broken English, but some very odd phrasing, malapropisms and incorrect use of grammar here and there. By V4 she's much less likely to make mistakes like that but still has features like relative lack of difference in stressed and unstressed syllables (resulting in a very "measured"-sounding prosody). I can see her having an almost exaggerated American intonation, almost as if she's reading a book aloud or something (reading a book aloud in an engaging way, not like the kid in school who reads it in the most flat and boring way possible).

    I think for Miku I might use lots of pitch bend to give her an almost "bouncing" intonation, but give her more measured, occasionally unsure or unsteady, stress, while Avanna's speech is flatter pitch-wise (nowhere near a monotone but much less pitch-bend-y) but has a more rhythmic, "lilting" stress pattern.

    Also, I tend to make Miku speak in a pretty high pitch which should be unsurprising (unless I'm using dark, solid, or V3 English where I pitch it down a few notches). Avanna, of course, speaks much lower (and making a Talkloid with Miku and switching it to Avanna produces some very jarring results; Avanna should not speak like a Japanese teen pop star, it's safe to say).
    mobius017
    You know a song we hardly ever hear in concerts? Iroha's "Moon." Awesome song, though it's a bit different, so getting the vibe to fit might be a bit of a challenge. Still, it'd be really cool to see that happen again!
    mobius017
    I'm feeling like a bacon breakfast sandwich, some coffee with too much sugar/etc. to be good for me, some sunshine, and a day off. Sound good to anyone else?
    mobius017
    kz had an incredible setlist at Digital Stars 2022. I go to kz for some of the sweetest classic Miku songs, but he can put together his work and others into a real banger setlist, too.
    mobius017
    After some research, awhile of failed trial and error experiments, and a period of having given up, I've just relearned that you can hold down the shift key and drag the mouse on the Presonus Presence instruments' keyboard to scroll to lower/higher notes.

    Why!? Why wasn't that more intuitive!? I mean, really, who would think to do that if they didn't already know they could? (I mean, granted, I thought of it at one point, but the fact that I was able to forget it and then struggle to find it again....)
    mobius017
    I'm real late to appreciate this, but it's amazing how much of a difference in pronunciation you can get by splitting a note. By shortening/lengthening vowels, it seems like you can sometimes change how they get pronounced; similar to consonants, you can also either make them more evident or less intrusive. The end result being, you can make vocals clearer, or at least get the pronunciation more like how you imagine it. (Sometimes, this work has made me wonder if I pronounce words correctly, or if I'm just making Miku speak more like I do.)

    Sometimes replacing phonemes doesn't do it, but taking explicit control of the lengths of your consonants and vowels does.
    • Like
    Reactions: Ronny777, TheStarPalace, MillyAqualine and 4 others
    Vector
    Vector
    I rely on that trick a lot. You can also use the Velocity parameter to adjust how prominent the consonant part is also, if they sound too intrusive. Similarly, I occasionally dip DYN down a bit when I can't get a vowel sound right. It just makes it sound less annunciated and can sort of hide weird pronunciation.
    AmazingStrange39
    AmazingStrange39
    English Vocaloids have a tendency to pronounce a word in kind of a short way? Like they're barely noticing the vowel or something- I've especially noticed this with Miku. So splitting the note in two helps me have Miku, for example, pronounce a word the way I want her to.
    Ronny777
    Ronny777
    That's is so true!! I do all my pitch bends this way. It's way more reliable for V5 than the drawing it.(i can never get a smooth transition) You can also add multiple different phonemes on a short/tiny note to drastically change the pronunciation of the following long note.
    mobius017
    I'm liking the new background on VocaDB. The other one had been up for awhile, and it makes a nice change!
    Ibis
    Ibis
    The rein of Sakura Miku has ended
    mobius017
    IIRC, aside from [Sil], there was at least one other Vocaloid phoneme that could be used to produce a silence/break between phonemes, right? Can someone remind me what it was?

    (Sadly, not sure which version/language it was in, if it existed. If there is such a thing, I'd be interested in knowing regardless!)
    • Like
    Reactions: WyndReed
    View previous comments…
    mobius017
    mobius017
    I'll give those a look, too!

    @peaches2217: You sort of answered this already, but I can't find [x] on the phoneme lists I've checked. Do you know where it comes from and if it had an intended usage besides being an alternative [Sil]?
    peaches2217
    peaches2217
    I honestly don’t remember. I think it’s just a matter of that phoneme not actually existing, so it makes the note silent.
    AmazingStrange39
    AmazingStrange39
    Miku's Japanese voicebanks and AVANNA and SONiKA all have [Asp] but unfortunately Miku English does not.
    mobius017
    Booster round 2!

    I haven't heard what anyone else's experience with the booster is, but aside from my arm hurting and maybe being marginally tired/fuzzy, I'm not experiencing any side effects. My side effects were never super bad to begin with, but I think it's at least plausible that my body has gotten better at handling this.
    mobius017
    I'm trying to make Miku hum. This is an interesting challenge.
    • Like
    Reactions: MillyAqualine, Ibis and WyndReed
    View previous comments…
    mobius017
    mobius017
    Hmm, maybe...? In that vein, I wonder if Growl would be useful.... Thank you!
    Vector
    Vector
    The other thing that comes to mind is putting exhalations in. I noticed when I was remixing that KIRA song that he has inhalation breath sounds all over the place when Miku is singing, which isn't something I've played with. (But I know Crypton supplied samples when I bought Miku V4X...) We tend to write humming as "hmm" in English because there's a breathy attack before the sustained humming sound. So maybe using a breath sample (which can be reversed in the DAW if they only have inhalation...) leading right into the hum?
    mobius017
    mobius017
    I've only played with adding breath effects a little in the past--basically just for an inhalation sound before a verse/long section of singing starts. That is a good idea to add one of those before humming.... I might give that a try.... Though, I dunno, I think my current approach with the h phoneme is actually producing pretty good results? Somehow I got rid of the harder h and m sounds, though I can't remember exactly how; at least partly, it's due to raising those lowered 1/64 notes on some of the hummed notes.

    Here's how it sounds currently:

    Here's a hopefully-understandable description of the notes (at least the first few of them). The h notes are all 1/64.
    [h u:] (lowered 1 semitone) [m] (vibrato: Fast Type 1, length: 88%)
    [h] [m] (accent: 0; vibrato: Fast Type 1, length: 100%)
    [h] [m] (accent: 0; vibrato: Fast Type 1, length: 100%)
    [h] (lowered 1 semitone) [m] (vibrato: Fast Type 1, length: 88%)

    (You might need to turn your volume up a little; it's kind of quiet.)
    mobius017
    That said, for lyrics that are more freeform/less uniformly metrical, actually getting the timing of the MIDI so the words sound correct can require a good bit of patience :) .
  • Loading…
  • Loading…
  • Loading…
  • Loading…
  • Loading…
  • Loading…
  • Loading…
Top Bottom