Datasets:

bookbot

/

ljspeech_phonemes

like 3

I simply used gruut to phonemize the text transcripts. The accuracy of the punctuation depends on their g2p implementation.

Is that mean I can use gruut tool to phonemize each text? Thanks! I thought you labeled the annotation manually.😁

You could use gruut, but it might not be always accurate because gruut uses American English (en-US), while your speech data might not be in that accent. Moreover, gruut can only work on the text level without any information from the speech data. This can be a problem if there are >1 pronunciations of the same word, e.g. the, a, homographs, etc., which can lead to inaccurate labels. So you might need to recheck for those cases.

As for manually annotated phoneme datasets, you can have a look at TIMIT. It is one sample of that.

extremely thankful！

Cheers