Files
storyteller-ml/tts/text/pipeline_components/phonetization/test_cmudict.py
T
2022-07-05 02:44:45 -04:00

16 lines
555 B
Python

from text.pipeline_components.phonetization.cmudict import CMUDict
from text.pipeline_components.phonetization.cmudict import DEFAULT_CMUDICT_PATH
class TestCmuDict:
def test_lookup_success(self):
cmudict = CMUDict(DEFAULT_CMUDICT_PATH)
arpabet_sequence = cmudict.lookup('word')
assert arpabet_sequence == ['W ER1 D'] # TODO: Split the phonemes!
def test_lookup_failure(self):
cmudict = CMUDict(DEFAULT_CMUDICT_PATH)
arpabet_sequence = cmudict.lookup('fakeword')
assert arpabet_sequence == None