G2p instead of espeak

[JRMeyer]

>>> geneing
[January 7, 2020, 2:33am]

ESPnet project uses g2p library for phoneme translation (available from
pip or https://github.com/Kyubyong/g2p).

I did a quick test and g2p appears to be more accurate than phonemizer
for my favorite semi-ambiguous sentences.

For example: slash
txt='Who's read the book.' slash
phonemize(txt,backend='espeak') slash
Out slash [56 slash ]: slash 'huːz ɹiːd ðə bʊk '

slash ' slash '.join(g2p(txt)) slash
Out slash [66 slash ]: 'HH UW1 EH1 S R EH1 D DH AH0 B UH1 K .'

Also, it's about 6x faster for a paragraph-long text: slash
%timeit phonemize(txt,backend='espeak') slash
195 ms ± 7.21 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

%timeit g2p(txt) slash
28.3 ms ± 261 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

[This is an archived TTS discussion thread from discourse.mozilla.org/t/g2p-instead-of-espeak]

[JRMeyer]

>>> erogol
[January 8, 2020, 3:53pm]

Thx for the recommendation. afaik it only supports english which would
prevent other languages. Maybe we can only enable it for Eng.

[Archived Post]