There are also other versions of similar tech like Cevio (https://cevio.jp https://cevio.jp) which can handle both singing and speech. It can sound a lot more r

There are also other versions of similar tech like Cevio (https://cevio.jp https://cevio.jp) which can handle both singing and speech. It can sound a lot more realistic and has the 'gimmick' of making voicebanks for various popular people in Japan. Often the differences in the software generated result compared to the person's voice are down to their tuning preferences rather than technical limitations (on a side note, Cevio uses RNNs while IIRC the original Vocaloids used to concatenate voice clippings and smooth out the transitions using some fourier space tricks, I'm not aware of how modern Vocaloid works but it wouldn't be too surprising if they too use deep learning). It initially became a thing in Japan because Japanese has fewer phonemes than most mainstream languages, so it was easier to make. Then it gained popularity for the cute characters and the freedom it gave to people who would go on to be music composers. Many popular Japanese composers got their start making and uploading Vocaloid songs on Youtube/NicoNico. On top of that, they have had a series of well designed rhythm games on several platforms for decades now which were pretty popular. These days they have a fairly popular mobile gacha/gambling game too. Nostalgia is probably also a big driving force since many people grew up with either the music or the games. These days it's still pretty popular within the anime fanbase outside Japan. Pre-Covid there used to be a vocaloid concert series every year which would alternate between Europe and US tours. The reason it hasn't gone too much more mainstream is likely that English and related languages have thousands of phonemes, so making a good voicebank is significantly harder.