def transcribe(self, output_dir, **kwargs):
orig_audio = os.path.join(output_dir, AUDIO_FILENAME)
resample_audio = os.path.join(output_dir, RESAMPLE_FILENAME)
transcript = os.path.join(output_dir, TEXT_FILENAME)
logging.info('Resampling audio file %s', orig_audio)
if gentle.resample(orig_audio, resample_audio) != 0:
logging.info('Failed to resample %s', orig_audio)
return -1
def on_progress(p):
for k,v in p.items():
logging.info('Transcribing %s, %s, %s', resample_audio, k, v)
logging.info('Starting to transcribe %s', output_dir)
with open(transcript, 'r', encoding='utf-8') as file:
text = file.read()
transcriber = gentle.ForcedAligner(self.resources, text, nthreads=self.nthreads, **kwargs)
output = transcriber.transcribe(resample_audio, progress_cb=on_progress, logging=logging)
logging.info('Finished transcribing %s', output_dir)
return output
class Transcriber():
def __init__(self, data_dir, nthreads=4, ntranscriptionthreads=2):
self.data_dir = data_dir
self.nthreads = nthreads
self.ntranscriptionthreads = ntranscriptionthreads
self.resources = gentle.Resources()
self.full_transcriber = gentle.FullTranscriber(self.resources, nthreads=ntranscriptionthreads)
self._status_dicts = {}
def get_status(self, uid):
return self._status_dicts.setdefault(uid, {})
def out_dir(self, uid):
return os.path.join(self.data_dir, 'transcriptions', uid)
# TODO(maxhawkins): refactor so this is returned by transcribe()
def next_id(self):
uid = None
while uid is None or os.path.exists(os.path.join(self.data_dir, uid)):
uid = uuid.uuid4().get_hex()[:8]
return uid
def transcribe(self, uid, transcript, audio, async, **kwargs):
status = self.get_status(uid)
status['status'] = 'STARTED'
output = {
'transcript': transcript
}
outdir = os.path.join(self.data_dir, 'transcriptions', uid)
tran_path = os.path.join(outdir, 'transcript.txt')
with open(tran_path, 'w') as tranfile:
tranfile.write(transcript)
audio_path = os.path.join(outdir, 'upload')
with open(audio_path, 'w') as wavfile:
wavfile.write(audio)
status['status'] = 'ENCODING'
wavfile = os.path.join(outdir, 'a.wav')
if gentle.resample(os.path.join(outdir, 'upload'), wavfile) != 0:
status['status'] = 'ERROR'
status['error'] = "Encoding failed. Make sure that you've uploaded a valid media file."
# Save the status so that errors are recovered on restart of the server
# XXX: This won't work, because the endpoint will override this file
with open(os.path.join(outdir, 'status.json'), 'w') as jsfile:
json.dump(status, jsfile, indent=2)
return
#XXX: Maybe we should pass this wave object instead of the
# file path to align_progress
wav_obj = wave.open(wavfile, 'r')
status['duration'] = wav_obj.getnframes() / float(wav_obj.getframerate())
status['status'] = 'TRANSCRIBING'
def on_progress(p):
for k,v in p.items():
status[k] = v
if len(transcript.strip()) > 0:
trans = gentle.ForcedAligner(self.resources, transcript, nthreads=self.nthreads, **kwargs)
elif self.full_transcriber.available:
trans = self.full_transcriber
else:
status['status'] = 'ERROR'
status['error'] = 'No transcript provided and no language model for full transcription'
return
output = trans.transcribe(wavfile, progress_cb=on_progress, logging=logging)
# ...remove the original upload
os.unlink(os.path.join(outdir, 'upload'))
# Save
with open(os.path.join(outdir, 'align.json'), 'w') as jsfile:
jsfile.write(output.to_json(indent=2))
with open(os.path.join(outdir, 'align.csv'), 'w') as csvfile:
csvfile.write(output.to_csv())
# Inline the alignment into the index.html file.
htmltxt = open(get_resource('www/view_alignment.html')).read()
htmltxt = htmltxt.replace("var INLINE_JSON;", "var INLINE_JSON=%s;" % (output.to_json()));
open(os.path.join(outdir, 'index.html'), 'w').write(htmltxt)
status['status'] = 'OK'
logging.info('done with transcription.')
return output
def transcribe(self, uid, transcript, audio, async_mode, **kwargs):
status = self.get_status(uid)
status['status'] = 'STARTED'
output = {'transcript': transcript}
outdir = os.path.join(self.data_dir, 'transcriptions', uid)
tran_path = os.path.join(outdir, 'transcript.txt')
with open(tran_path, 'w') as tranfile:
tranfile.write(transcript)
if not isinstance(audio, str):
audio_path = os.path.join(outdir, 'upload')
with open(audio_path, 'wb') as wavfile:
wavfile.write(audio)
status['status'] = 'ENCODING'
wavfile = os.path.join(outdir, 'a.wav')
# if ((not isinstance(audio, str)) and gentle.resample(os.path.join(outdir, 'upload'), wavfile) != 0) or gentle.resample(audio, wavfile) != 0:
if (not isinstance(audio, str)) and gentle.resample(
os.path.join(outdir, 'upload'), wavfile) != 0:
status['status'] = 'ERROR'
status[
'error'] = "Encoding failed. Make sure that you've uploaded a valid media file."
# Save the status so that errors are recovered on restart of the server
# XXX: This won't work, because the endpoint will override this file
with open(os.path.join(outdir, 'status.json'), 'w') as jsfile:
json.dump(status, jsfile, indent=2)
return
if isinstance(audio, str) and gentle.resample(audio, wavfile) != 0:
status['status'] = 'ERROR'
status[
'error'] = "Encoding failed. Make sure that you've referenced a valid media URL."
# Save the status so that errors are recovered on restart of the server
# XXX: This won't work, because the endpoint will override this file
with open(os.path.join(outdir, 'status.json'), 'w') as jsfile:
json.dump(status, jsfile, indent=2)
return
# XXX: Maybe we should pass this wave object instead of the
# file path to align_progress
if not isinstance(audio, str):
wav_obj = wave.open(wavfile, 'rb')
status['duration'] = wav_obj.getnframes() / \
float(wav_obj.getframerate())
status['status'] = 'TRANSCRIBING'
def on_progress(p):
print(p)
for k, v in p.items():
status[k] = v
if len(transcript.strip()) > 0:
trans = gentle.ForcedAligner(self.resources,
transcript,
nthreads=self.nthreads,
**kwargs)
elif self.full_transcriber.available:
trans = self.full_transcriber
else:
status['status'] = 'ERROR'
status[
'error'] = 'No transcript provided and no language model for full transcription'
return
output = trans.transcribe(wavfile,
progress_cb=on_progress,
logging=logging)
# ...remove the original upload
if not isinstance(audio, str):
os.unlink(os.path.join(outdir, 'upload'))
# Save
with open(os.path.join(outdir, 'align.json'), 'w') as jsfile:
jsfile.write(output.to_json(indent=2))
with open(os.path.join(outdir, 'align.csv'), 'w') as csvfile:
csvfile.write(output.to_csv())
# Inline the alignment into the index.html file.
htmltxt = open(get_resource('www/view_alignment.html')).read()
htmltxt = htmltxt.replace("var INLINE_JSON;",
"var INLINE_JSON=%s;" % (output.to_json()))
open(os.path.join(outdir, 'index.html'), 'w').write(htmltxt)
status['status'] = 'OK'
logging.info('done with transcription.')
return output
class Transcriber():
def __init__(self, data_dir, nthreads=4, ntranscriptionthreads=2):
self.data_dir = data_dir
self.nthreads = nthreads
self.ntranscriptionthreads = ntranscriptionthreads
self.resources = gentle.Resources()
self.full_transcriber = gentle.FullTranscriber(
self.resources, nthreads=ntranscriptionthreads)
self._status_dicts = {}
def get_status(self, uid):
return self._status_dicts.setdefault(uid, {})
def out_dir(self, uid):
return os.path.join(self.data_dir, 'transcriptions', uid)
# TODO(maxhawkins): refactor so this is returned by transcribe()
def next_id(self):
uid = None
while uid is None or os.path.exists(os.path.join(self.data_dir, uid)):
uid = uuid.uuid4().get_hex()[:8]
return uid
def transcribe(self, uid, transcript, audio, async, **kwargs):
status = self.get_status(uid)
status['status'] = 'STARTED'
output = {'transcript': transcript}
outdir = os.path.join(self.data_dir, 'transcriptions', uid)
tran_path = os.path.join(outdir, 'transcript.txt')
with open(tran_path, 'w') as tranfile:
tranfile.write(transcript)
audio_path = os.path.join(outdir, 'upload')
with open(audio_path, 'w') as wavfile:
wavfile.write(audio)
status['status'] = 'ENCODING'
wavfile = os.path.join(outdir, 'a.wav')
if gentle.resample(os.path.join(outdir, 'upload'), wavfile) != 0:
status['status'] = 'ERROR'
status[
'error'] = "Encoding failed. Make sure that you've uploaded a valid media file."
# Save the status so that errors are recovered on restart of the server
# XXX: This won't work, because the endpoint will override this file
with open(os.path.join(outdir, 'status.json'), 'w') as jsfile:
json.dump(status, jsfile, indent=2)
return
#XXX: Maybe we should pass this wave object instead of the
# file path to align_progress
wav_obj = wave.open(wavfile, 'r')
status['duration'] = wav_obj.getnframes() / float(
wav_obj.getframerate())
status['status'] = 'TRANSCRIBING'
def on_progress(p):
for k, v in p.items():
status[k] = v
if len(transcript.strip()) > 0:
trans = gentle.ForcedAligner(self.resources,
transcript,
nthreads=self.nthreads,
**kwargs)
elif self.full_transcriber.available:
trans = self.full_transcriber
else:
status['status'] = 'ERROR'
status[
'error'] = 'No transcript provided and no language model for full transcription'
return
output = trans.transcribe(wavfile,
progress_cb=on_progress,
logging=logging)
# ...remove the original upload
os.unlink(os.path.join(outdir, 'upload'))
# Save
with open(os.path.join(outdir, 'align.json'), 'w') as jsfile:
jsfile.write(output.to_json(indent=2))
with open(os.path.join(outdir, 'align.csv'), 'w') as csvfile:
csvfile.write(output.to_csv())
# add file datas
sens_end_index = trans.ms.get_sentences_index()
res = output.to_json()
res = json.loads(res, encoding='utf-8', strict=True)
time_sentences_index = []
ss_dot = 0
s_pos = None
time_pos = 0
try:
for i, w in enumerate(res['words']):
if w["case"] != "success":
continue
end_v = w['endOffset']
start_v = w['startOffset']
if s_pos is None:
s_pos = start_v
time_pos = i
if end_v >= sens_end_index[ss_dot]:
ss_dot += 1
time_sentences_index.append(
(res['words'][time_pos]["start"],
res['words'][i]["end"]))
time_pos = i
s_pos = end_v
if len(sens_end_index) != len(time_sentences_index):
time_sentences_index.append(
(res['words'][time_pos]["start"], res['words'][-1]["end"]))
#print sens_end_index, len(sens_end_index)
#print time_sentences_index, len(time_sentences_index)
sens_str = trans.ms.get_sentences_string()
save_ss = ""
for i, t in enumerate(time_sentences_index):
#print "{{time}}%s/%s{{end}}" % (str(round(float(t[0]), 2)), str(round(float(t[1]), 2)))
#print "{{raw}}%s{{end}}" % (str(sens_str[i]))
save_ss += "{{time}}" + str(round(float(t[0]), 2)) + "/" + str(
round(float(t[1]), 2)) + "{{end}}\n"
save_ss += "{{raw}}" + sens_str[i] + "{{end}}\n"
with open(os.path.join(outdir, 'time.csv'), 'w') as timefile:
timefile.write(save_ss)
except Exception as e:
print traceback.format_exc()
# Inline the alignment into the index.html file.
htmltxt = open(get_resource('www/view_alignment.html')).read()
htmltxt = htmltxt.replace("var INLINE_JSON;",
"var INLINE_JSON=%s;" % (output.to_json()))
open(os.path.join(outdir, 'index.html'), 'w').write(htmltxt)
status['status'] = 'OK'
logging.info('done with transcription.')
return output