I have been testing Saki AI (full version), and I believe that there may be some phonemes missing on the dataset. For instance, any syllable starting with “my” makes a sound that resembles an “f”. Syllables starting with “hy” don’t sound correct either. I don’t know how to describe “ry”, but those do not sound correct either. Those are the most prominent cases, but I found some issues involving most iotized phonemes.
I am attaching an audio sample with the corresponding svp file that show a few cases where I feel that the phonemes don’t sound right.
phonemes.zip (1,1 MB)
「いいね！」 1
Yeah, I’m having similar problems. Saki AI seems to struggle with palatalized phonemes in general. For a lot of them she only pronounces it some of the time, and other times she can’t pronounce it at all. Note that I am using the Lite version.
“ky” seems to mostly work, though it has some odd consonant length.
“gy” has the same problems as “ky”.
“ty” is just all kinds of messy; it sounds like there are samples recorded for it, but there’s an odd fuzziness in the phoneme that makes it borderline unintelligible.
“dy” seems to work before “a” and “u”, but omits the “y” preceding other vowels.
“by” only seems to work preceding an “a”, and even then it sounds somewhat strange.
“py” doesn’t seem to work properly at all.
“my” doesn’t work at all.
“ny” seems to mostly work, though has odd consonant length and some fuzziness.
“ry”… I don’t even know what’s going on with that sound.
“hy” seems to mostly work, but has odd consonant length, and omits the “y” preceding “o”.