Mis-pronunciation of non-standard numbers about rhasspy HOT 8 OPEN

Which text to speech system are you using?

from rhasspy.

Larynx.

from rhasspy.

That sounds like the appropriate gruut package may not be installed then.

from rhasspy.

Hmmmm, how would I check? I installed the rhasspy version that comes as an HA add on. It should be whatever the default is for that.

from rhasspy.

Found another interesting tidbit digging in the logs. It looks like numbers are being replaced by '#' signs. Using a random sentence to test ordinal numbers and what not I get the following:

[DEBUG:2023-03-27 19:34:02,522] larynx: Phonemes for 'Today is the 27th 1st 2nd 23rd 7': ['#', 't', 'ə', 'd', 'ˈ', 'eɪ', '#', 'ˈ', 'ɪ', 'z', '#', 'ð', 'ə', '#', 'θ', '#', 's', 't', '#', 'n', 'd', '#', 'ɹ', 'd', '#', 's', 'ˈ', 'ɛ', 'v', 'ə', 'n', '#', '‖', '‖']

Note at the end that the seven is translated properly, but the ordinal numbers get turned into '#'s for some reason.

from rhasspy.

Sorry, should have noted that right above that i see this:

[DEBUG:2023-03-27 19:34:00,716] gruut.phonemize: Guessing pronunciations for {'27th'}
[DEBUG:2023-03-27 19:34:00,719] phonetisaurus: ['phonetisaurus-apply', '--model', '/usr/lib/rhasspy/.venv/lib/python3.7/site-packages/gruut/data/en-us/g2p.fst', '--word_list', '/tmp/tmpccme_ugu.txt', '--nbest', '1']
[DEBUG:2023-03-27 19:34:02,522] larynx: Words for 'Today is the 27th 1st 2nd 23rd 7': ['today', 'is', 'the', '27th', '1st', '2nd', '23rd', 'seven']

So it looks like it's attempting to do SOMETHING with gruut.

from rhasspy.

The phonetisaurus model doesn't contain digits, so it's dropping those and just guessing the pronunciation of "th" and "st" for "27th" and "1st", etc.

Newer version of gruut can handle this, but Rhasspy 2.5.11 may be on the older one.

from rhasspy.