def _convert(self, audio):
#import speech_recognition as sr
with audio.get_filename() as filename:
with sr.AudioFile(filename) as source:
clip = self.recognizer.record(source)
_json = self._query_api(clip)
if 'results' in _json:
results = _json['results']
else:
raise Exception(
'received invalid results from API: {0}'.format(str(_json)))
elements = []
for result in results:
if result['final'] is True:
timestamps = result['alternatives'][0]['timestamps']
if self.resolution is 'words':
for entry in timestamps:
text = entry[0]
start = entry[1]
end = entry[2]
elements.append(TextStim(text=text, onset=start,
duration=end-start))
elif self.resolution is 'phrases':
text = result['alternatives'][0]['transcript']
start = timestamps[0][1]
end = timestamps[-1][2]
elements.append(TextStim(text=text, onset=start,
duration=end-start))
return ComplexTextStim(elements=elements)
def _convert(self, audio):
import speech_recognition as sr
with sr.AudioFile(audio.filename) as source:
clip = self.recognizer.record(source)
text = getattr(self.recognizer, self.recognize_method)(clip,
self.api_key)
return ComplexTextStim(text=text)
def _convert(self, audio):
verify_dependencies(['sr'])
with audio.get_filename() as filename:
with sr.AudioFile(filename) as source:
clip = self.recognizer.record(source)
text = getattr(self.recognizer, self.recognize_method)(clip, self.api_key)
return ComplexTextStim(text=text)
def _convert(self, audio):
import speech_recognition as sr
with sr.AudioFile(audio.filename) as source:
clip = self.recognizer.record(source)
result = self._query_api(clip)
timestamps = result['results'][0]['alternatives'][0]['timestamps']
elements = []
for i, entry in enumerate(timestamps):
elements.append(
TextStim(text=entry[0],
onset=entry[1],
duration=entry[2] - entry[1]))
return ComplexTextStim(elements=elements)
def getVector(self, stim, cbow=False):
if not isinstance(stim, list):
stim = [stim]
if cbow == True:
stim_cbows = self.generateCbows(stim)
stim = stim_cbows
'''
We only implement average word embedding'
'''
'''we already have the embeddings loaded '''
num_dims = self.wvModel.vector_size
'''the stimuli is a list '''
''' we create average embedding.
Need to decide whether move the functionality
to an util class later
'''
embedding_average_vectors = []
for s in stim:
complex_s = ComplexTextStim(text=s.lower())
embeddings = self.transform(complex_s)
embedding_average_vector = np.zeros(num_dims)
numWords = 0
for embedding in embeddings:
if self.content_only == True:
if embedding.stim.data in self.stop_words or \
not embedding.stim.data.isalnum() :
continue
for index, value in enumerate(embedding._data[0]):
embedding_average_vector[index] += value
numWords += 1
if numWords > 0:
for index in range(num_dims):
embedding_average_vector[index] /= numWords
embedding_average_vectors.append(embedding_average_vector)
features = ['%s%d' % (self.prefix, i) for i in range(num_dims)]
return ExtractorResult(embedding_average_vectors,
stim,
self,
features=features)
def _convert(self, stim):
request = self._build_request(stim)
response = self._query_api(request)
if 'error' in response:
raise Exception(response['error']['message'])
words = []
if 'results' in response:
for result in response['results']:
transcription = result['alternatives'][0]
for w in transcription['words']:
onset = float(w['startTime'][:-1])
duration = float(w['endTime'][:-1]) - onset
words.append(TextStim(text=w['word'],
onset=onset,
duration=duration))
return ComplexTextStim(elements=words)
def _convert(self, audio):
verify_dependencies(['rev_ai'])
msg = "Beginning audio transcription with a timeout of %fs. Even for "\
"small audios, full transcription may take awhile." % self.timeout
logging.warning(msg)
if audio.url:
job = self.client.submit_job_url(audio.url)
else:
with audio.get_filename() as filename:
job = self.client.submit_job_local_file(filename)
operation_start = time.time()
response = self.client.get_job_details(job.id)
while (response.status == rev_ai.JobStatus.IN_PROGRESS) and \
(time.time() - operation_start) < self.timeout:
response = self.client.get_job_details(job.id)
time.sleep(self.request_rate)
if (time.time() - operation_start) >= self.timeout:
msg = "Conversion reached the timeout limit of %fs." % self.timeout
logging.warning(msg)
if response.status == rev_ai.JobStatus.FAILED:
raise Exception('API failed: %s' % response.failure_detail)
result = self.client.get_transcript_object(job.id)
elements = []
order = 0
for m in result.monologues:
for e in m.elements:
if e.type_ == 'text':
start = e.timestamp
end = e.end_timestamp
elements.append(
TextStim(text=e.value,
onset=start,
duration=end - start,
order=order))
order += 1
return ComplexTextStim(elements=elements, onset=audio.onset)
def _convert(self, audio):
verify_dependencies(['sr'])
offset = 0.0 if audio.onset is None else audio.onset
with audio.get_filename() as filename:
with sr.AudioFile(filename) as source:
clip = self.recognizer.record(source)
_json = self._query_api(clip)
if 'results' in _json:
results = _json['results']
else:
raise Exception('received invalid results from API: {0}'.format(
str(_json)))
elements = []
order = 0
for result in results:
if result['final'] is True:
timestamps = result['alternatives'][0]['timestamps']
if self.resolution is 'words':
for entry in timestamps:
text = entry[0]
start = entry[1]
end = entry[2]
elements.append(
TextStim(text=text,
onset=offset + start,
duration=end - start,
order=order))
order += 1
elif self.resolution is 'phrases':
text = result['alternatives'][0]['transcript']
start = timestamps[0][1]
end = timestamps[-1][2]
elements.append(
TextStim(text=text,
onset=offset + start,
duration=end - start,
order=order))
order += 1
return ComplexTextStim(elements=elements, onset=audio.onset)
