I've been using Translate to help a bit with Japanese learning. What I discovered just now is that the text to speech pronunciation doesn't seem to operate correctly for Japanese when using kanji? Maybe someone more experienced can confirm.
For example: 貴金属 the hiragana form is spoken correctly ききんぞく (kikinzoku), but the kanji sounds what I'm guessing is some Chinese reading, although the text is correctly automatically identified as Japanese. The romanization of the kanji is also correct, so it's just whatever is happening with the speech API you're using.
Sorry if this has been mentioned already, this thread is really long!