I suppose it means per speaker. And it is based on a simplified style tts 2 which from my small dive into the subject seems one of the smaller models achieving great quality.
in the linked audio sample it says the training data is mostly english. also another comment claims that the japanese quality is not good, so i'd be suspicious about all the other languages.
I suppose it means per speaker. And it is based on a simplified style tts 2 which from my small dive into the subject seems one of the smaller models achieving great quality.
Yeah that surprised me as well - seems low vs what is used on text llms . To be fair 100 hours of speaking is a lot of speaking though
But it covers five? Languages so if all equal it’s just 20 hours per language.
in the linked audio sample it says the training data is mostly english. also another comment claims that the japanese quality is not good, so i'd be suspicious about all the other languages.