vocaloid data rummaging, part 2/?: a thousand voicebanks

this is part 2 in a series of posts exploring vocaloid song data aggregated from vocaDB and niconico. you can read part 1 here.

issen kounen may not register as the most popular vocal synth song this year, or even the PV with the most featured voicebanks this year to break into the vocaloid hall of fame, but it may very well be the most uniquely representative palette of voice banks to be featured in any modern vocal synth song. even while that can be true, miku steals attention as she wants to and commands visual focus in the PV, leading us along an endless run out into an unseen distance. vocally, she’s only the supporting hum behind a full ensemble of characters: four vocaloids. three CeVIO voicebanks. two synthV voicebanks from competing labels. there’s even an UTAU here that has only registered use in a little over 250 songs. while they all sing about a future stretching out 1000 light years, still a blink of the eye from certain vantage points, it only took us about 15 years to get to a point where they would all sing together.

as I often ground things to when speaking about the fandom broadly, miku is, to most people, both the genesis and totality of vocal synth music. there is a core truth to that when no voicebank has so far come close to her individual influence, both in creation and consumption. in much the same way, you’ll also find that fans like me will often code switch between using “vocaloid” and “vocal synths” when talking about the scene in total, as they were at one point entirely equivalent. as we examine the introduction of voicebanks, we’ll see why the story is richer than either outdated understanding and reveals a deeper shift in how people have come to appreciate vocal synths.

A quick note on Miku being first — sort of

any visualizations we try to generate capturing vocalist data reveal the obvious truth that miku has been a runaway success. since we are looking at only niconico data here, which essentially discards everything preceding 2007, it’s important to consider that miku was really one lucky combination of successes converging on her release. not only was she not crypton’s first foray into vocaloid, but she was certainly not close to being even the first vocal synth to gain mindshare among japanese music creatives. while even earlier groups like kraftwerk developed a sound from lifting vocoders out of research labs and on to LPs, those uses didn’t fully represent generated vocals like what became possible with the release of vocalwriter. vocalwriter is by some accounts (mostly scrounged from old catalogs) the first commercial vocal music synthesizer, released for the macintosh in 1998 featuring a variety of voice options that let you assign lyrics to MIDI tones in effectively the same method as vocaloid would later develop. seiya-murai of bemani fame defined a style primarily around his use of vocalwriter, most notably using sarah, tracy, and andy’s voices for his trio of pop’n characters, and it also found itself wormed into other game contexts like katamari damacy. while these voices lack fidelity, they have a basic charm that keeps them safely away from the cliff of the uncanny or overambitious. miku certainly achieved a more convincing technical fidelity, but I like to remind myself as we look at her popularity that the initial success she experienced had as much to do with the networking effects emerging across a nascent social video boom as any direct interest that came from the development of vocal synths. a few individuals like murai are always exploring the depths, and others too are able to play effective tastemakers (which we’ll explore a little later), but the changes across years we see in vocalist usage are more likely to be a representation of creatives coalescing around what’s popular rather than a cognizant desire to explore alternatives. hideki matsutake handily proved meiko’s and kaito’s worth in 2003, but how many people were showing up to hear it tucked away on a compilation album? miku was “first” in most senses to most people, and for that reason will probably always exist overarching any other cultural changes I might make note of here.

Producers look to popularity and capabilities

the chart at the top of this post backs up what is probably fairly intuitive to even casual observers: miku is hugely popular, producers didn’t take long to build up other options, and excitement around those options drove the creation of more songs. these numbers do differ from the overall song numbers seen in my first post, since we are counting multiple vocalists for one song, but those trends we noted there still show up here, and this granular view more obviously highlights some of the shifts that occurred in the scene. GUMI was an early example of a commercial voicebank outside of the cryptonloid dynasty, released in 2009 by internet co ltd, and she was rewarded for this by also experiencing steady natural growth alongside other available voicebanks. while it is tempting to view her as a scrappy upstart going up against names with larger starpower¹, she was in reality eclipsing every crypton voicebank not named miku in use at her peak in 2012. much of vocaloid’s early popularity was informed by the character designs being a great vector for art and music, and so having defined, popular characters to fall back on was especially convenient for creatives getting acclimated to a new kind of collaborative medium.

leading into 2016, GUMI’s usage started declining without any serious replacement to fill the void she was leaving behind. the standard vocaloids that grew out the scene were beginning to contract as well, with miku as a sort of bellwether as her usage began slipping during the years that GUMI took shape. while the reasons why vocaloid overall was on a downturn were the focus of fan hysterics of the time, the honest way to view this decline is that vocaloid was cooling off from a fad phase. new commercial entrants were still appearing, but producers were either abandoning the scene entirely or now trying to “legitimize” themselves outside of a scrappy doujin context. vocaloid’s tech backbone was also growing stagnant, and in fact many of the expansions that were being developed for the vocaloid 2 (V2) engine were seemingly blindsided by the announcement of a V3 engine and instead delayed their release in favor of a rewrite. diehards and famous labels like EXIT TUNES did champion emerging options like mayu and seeU, and miku did eventually manage a V3 release with english support, but these releases failed to capture enduring attention, or in miku’s case even initially lacked voices from her V2 counterparts. there are undoubtedly more confounding factors to consider here, like shifting fan tastes fanning out to other franchises, but the realization that no new voicebanks were able to emerge in significant proportion at this time I would at least partially attribute to commercial providers not living up to creator expectations and also lagging to capitalize on a boom period. V3 voicebanks delivered only incremental improvements, V4 delivered effectively none at all, and V5 caused most companies to transition away from developing voicebanks for the vocaloid engine altogether. with no real reason to upgrade to new releases, producers continued with tested and still popular voicebanks that were now saturating nico, or were instead left to probe options like UTAU that always existed on the periphery of vocaloid.

“They had a community that wanted a more versatile engine with a greater variety of voices, better English support, more intuitive tuning, an update to XSY, more support for other languages like Korean and Spanish and any support at all for French. People wanted something substantial. And with new leadership, maybe this could happen, right? Nope.”
_{— From a Reddit oral history examining the decline of VOCALOID}

Branching out revives a flagging scene

many producers would never stop exploring other voicebanks. during 2016, our grouping of “other” vocalists not represented by the listed heavy hitters managed a quarter of total usage — the only point at which a category besides miku was able achieve this level of use. while “other” is a catch-all ranging from someone’s artisan UTAU to a commercial voicebank that may as well have never been sold, this is the clearest picture yet that producers were cooling off of the crypton voicebanks that had effectively defined the scene, while synth otaku were still exploring their options. since the overall number of songs being published was declining, it’s notable that the “other” grouping, along with miku, were the only slices to float proportional growth during this time. while again it’s more appropriate to attribute the majority of that to enduring use of UTAU as a free and open alternative, the full timeline makes clear that crypton, and vocaloid by extension, was succeeding off of inertia more than anything else.

vocal synths found a second wind following this decline when engines like synthV, CeVIO, and NEUTRINO disrupted the scene. the groundswell of voicebanks like kiritan coincides with engines that began offering truly unique functionality that vocaloid failed to ever develop, including more human-like tuning made possible by neural networks. even when they do experience natural growth and decline in their usage, much like GUMI and all other voicebanks do, we now see newer releases hitting the ground running and proving that they’re able to effectively capitalize on the energy producers have for new options. the sheer growth of new songs being made makes this especially obvious: even though many voicebanks exist proportionally on the margins, they represent significant value to the scene’s overall growth when they can register published songs exceeding 500 in a year. where KAFU effectively replaces the usage kiritan once commanded in 2020, kiritan can settle into company alongside the collection of voicebanks like chis-A and SEKAI that maintain microcosms of influence and dedicated producers that adopt them as part of their identities.

vocaloid in its early history saw crypton characters molded more as idols rather than instruments, as vocal synths are now more colloquially aligned, with an array of mix media offerings like the project diva games and concerts developed by sega. as vocal synth music has emerged from a low point, while it’s clear that crypton has continued to lean further into this approach with efforts like project sekai, much of the core vocal synth community has reoriented itself around the producers themselves rather than the vocalist characters they feature. this, in turn, has given producers more freedom to adopt new sounds and further niches. while allback instead feels like a nostalgic throwback to this prior era full of character PVs, the seeds for it were sown more recently by inabakumori, who effectively raised yuki out of irrelevance by using her exclusively around 2018. with lagtrain as their smash hit, yuki’s usage has now risen almost every year since 2018, and current trends make it likely we will see a repeat of the same for 2023. allback is poised to eclipse that inabakumori influence, given the attention it’s receiving from vtubers and tiktokers alike, but it’s another good reminder that trying to gauge which voicebanks will establish popularity is like trying to read the tea leaves — teto is now a synthV and una is getting upgraded for V6, after all — and that it’s more valuable to step back and appreciate that the scene is probably the healthiest it’s ever been.