def test_new_transcription_using_use_existing(tmpdir):
"""
Using the use_existing when an existing transcription does not exist is okay.
"""
kaldi = KaldiInterface(f'{tmpdir}/state')
kaldi.new_transcription('transcription_w', use_existing=True)
return
def test_override_and_use_existing(tmpdir):
"""
Cannot have both the override and use_existing parameters set to True.
"""
kaldi = KaldiInterface(f'{tmpdir}/state')
with pytest.raises(ValueError):
kaldi.new_transcription('transcription_w',
override=True,
use_existing=True)
return
def test_two_new_transcription_with_same_name(tmpdir):
"""
Trying to create two transcriptions with the same name without override or
use_existing set to True will produce a ValueError.
"""
kaldi = KaldiInterface(f'{tmpdir}/state')
kaldi.new_transcription('transcription_w')
with pytest.raises(ValueError):
kaldi.new_transcription('transcription_w')
return
def test_existing_transcription_using_override(tmpdir):
"""
Use override to delete a transcription with the same name and create a totally
new transcription with the same name.
"""
kaldi = KaldiInterface(f'{tmpdir}/state')
t1 = kaldi.new_transcription('transcription_w')
t1_hash = t1.hash
t2 = kaldi.new_transcription('transcription_w', override=True)
# note t1 can no longer be used
assert len(kaldi.list_transcriptions()) == 1
assert t1_hash != t2.hash
return
def test_error_when_writing_to_protected_property(tmpdir):
"""
An error is raised when there is an attempt to write to a protected
property.
"""
kaldi = KaldiInterface(f'{tmpdir}/state')
t = kaldi.new_transcription('transcription_w')
with pytest.raises(AttributeError):
t.has_been_transcribed = True
with pytest.raises(AttributeError):
t.exporter = "some obj"
with pytest.raises(AttributeError):
t.state = "Not a valid"
def test_new_transcription(tmpdir):
"""
Check the state of a new transcription.
"""
kaldi = KaldiInterface(f'{tmpdir}/state')
t = kaldi.new_transcription('transcription_w')
assert t.has_been_transcribed == False
assert t.exporter == None
assert t.state == json.loads(f"""
{{
"name": "transcription_w",
"hash": "{t.hash}",
"date": "{t.date}",
"model": null,
"has_been_transcribed": false,
"exporter": null
}}
""")
return
def test_new_transcription_using_override(tmpdir):
"""
Using override has no effect when the pron dict with the same name does not
exist.
"""
kaldi = KaldiInterface(f'{tmpdir}/state')
t = kaldi.new_transcription('transcription_w', override=True)
assert t.has_been_transcribed == False
assert t.exporter == None
assert t.state == json.loads(f"""
{{
"name": "transcription_w",
"hash": "{t.hash}",
"date": "{t.date}",
"model": null,
"has_been_transcribed": false,
"exporter": null
}}
""")
return
ds.add_directory('/recordings/transcribed', filter=['eaf', 'wav'])
ds.process()
# Step 2
# ======
# Build pronunciation dictionary
pd = kaldi.new_pron_dict('pd')
pd.link(ds)
pd.set_l2s_path('/recordings/letter_to_sound.txt')
pd.generate_lexicon()
# Step 3
# ======
# Link dataset and pd to a new model, then train the model.
m = kaldi.new_model('mx')
m.link(ds, pd)
m.build_kaldi_structure()
m.train() # may take a while
# Step 4
# ======
# Make a transcription interface and transcribe unseen audio to elan.
t = kaldi.new_transcription('tx')
t.link(m)
with open('/recordings/untranscribed/audio.wav', 'rb') as faudio:
t.prepare_audio(faudio)
# t.transcribe_align()
t.transcribe()
# print(t.elan().decode('utf-8'))
print(t.text().decode('utf-8'))