def serve(port=8765, interface='0.0.0.0', installSignalHandlers=0, nthreads=4, ntranscriptionthreads=2, data_dir=get_datadir('webdata')):
logging.info("SERVE %d, %s, %d", port, interface, installSignalHandlers)
if not os.path.exists(data_dir):
os.makedirs(data_dir)
zip_dir = os.path.join(data_dir, 'zip')
if not os.path.exists(zip_dir):
os.makedirs(zip_dir)
f = File(data_dir)
f.putChild(b'', File(get_resource('www/index.html')))
f.putChild(b'status.html', File(get_resource('www/status.html')))
f.putChild(b'preloader.gif', File(get_resource('www/preloader.gif')))
trans = Transcriber(data_dir, nthreads=nthreads, ntranscriptionthreads=ntranscriptionthreads)
trans_ctrl = TranscriptionsController(trans)
f.putChild(b'transcriptions', trans_ctrl)
trans_zippr = TranscriptionZipper(zip_dir, trans)
f.putChild(b'zip', trans_zippr)
s = Site(f)
logging.info("about to listen")
reactor.listenTCP(port, s, interface=interface)
logging.info("listening")
reactor.run(installSignalHandlers=installSignalHandlers)
def serve(port=8765, interface='0.0.0.0', installSignalHandlers=0, nthreads=4, ntranscriptionthreads=2, data_dir=get_datadir('webdata')):
logging.info("SERVE %d, %s, %d", port, interface, installSignalHandlers)
if not os.path.exists(data_dir):
os.makedirs(data_dir)
zip_dir = os.path.join(data_dir, 'zip')
if not os.path.exists(zip_dir):
os.makedirs(zip_dir)
f = File(data_dir)
f.putChild('', File(get_resource('www/index.html')))
f.putChild('status.html', File(get_resource('www/status.html')))
f.putChild('preloader.gif', File(get_resource('www/preloader.gif')))
trans = Transcriber(data_dir, nthreads=nthreads, ntranscriptionthreads=ntranscriptionthreads)
trans_ctrl = TranscriptionsController(trans)
f.putChild('transcriptions', trans_ctrl)
trans_zippr = TranscriptionZipper(zip_dir, trans)
f.putChild('zip', trans_zippr)
s = Site(f)
logging.info("about to listen")
reactor.listenTCP(port, s, interface=interface)
logging.info("listening")
reactor.run(installSignalHandlers=installSignalHandlers)
def serve(port=8765, interface='0.0.0.0', installSignalHandlers=0, nthreads=4, ntranscriptionthreads=2, data_dir=get_datadir('webdata'), modelDir='exp'):
logging.info("SERVE %d, %s, %d", port, interface, installSignalHandlers)
if not os.path.exists(data_dir):
os.makedirs(data_dir)
zip_dir = os.path.join(data_dir, 'zip')
if not os.path.exists(zip_dir):
os.makedirs(zip_dir)
f = File(data_dir)
f.putChild('', File(get_resource('www/index.html')))
f.putChild('status.html', File(get_resource('www/status.html')))
f.putChild('preloader.gif', File(get_resource('www/preloader.gif')))
resources = gentle.Resources(modelDir)
trans = Transcriber(data_dir, nthreads=nthreads, ntranscriptionthreads=ntranscriptionthreads, modelDir=modelDir)
config = trans.config
logging.info("CONFIG: samplerate %d, silencephones %s, context-width %s", config['samplerate'], config['silencephones'], config['context-width'])
trans_ctrl = TranscriptionsController(trans)
f.putChild('transcriptions', trans_ctrl)
trans_zippr = TranscriptionZipper(zip_dir, trans)
f.putChild('zip', trans_zippr)
s = Site(f)
logging.info("about to listen")
reactor.listenTCP(port, s, interface=interface)
logging.info("listening")
reactor.run(installSignalHandlers=installSignalHandlers)
def render_POST(self, req):
uid = self.transcriber.next_id()
tran = req.args.get(b'transcript', [b''])[0].decode()
audio = req.args[b'audio'][0]
disfluency = True if b'disfluency' in req.args else False
conservative = True if b'conservative' in req.args else False
kwargs = {
'disfluency': disfluency,
'conservative': conservative,
'disfluencies': set(['uh', 'um'])
}
async = True
if b'async' in req.args and req.args[b'async'][0] == b'false':
async = False
# We need to make the transcription directory here, so that
# when we redirect the user we are sure that there's a place
# for them to go.
outdir = os.path.join(self.transcriber.data_dir, 'transcriptions', uid)
os.makedirs(outdir)
# Copy over the HTML
shutil.copy(get_resource('www/view_alignment.html'),
os.path.join(outdir, 'index.html'))
result_promise = threads.deferToThreadPool(reactor,
reactor.getThreadPool(),
self.transcriber.transcribe,
uid, tran, audio, async,
**kwargs)
if not async:
def write_result(result):
'''Write JSON to client on completion'''
req.setHeader("Content-Type", "application/json")
req.setHeader("X-Transcription-Id", uid)
req.write(result.to_json(uid=uid, indent=2).encode())
req.finish()
result_promise.addCallback(write_result)
result_promise.addErrback(lambda _: None) # ignore errors
req.notifyFinish().addErrback(lambda _: result_promise.cancel())
return NOT_DONE_YET
req.setResponseCode(FOUND)
req.setHeader(b"Location", "/transcriptions/%s" % (uid))
return b''
def render_POST(self, req):
uid = self.transcriber.next_id()
tran = req.args.get(b'transcript', [b''])[0].decode()
audio = req.args[b'audio'][0]
disfluency = True if b'disfluency' in req.args else False
conservative = True if b'conservative' in req.args else False
kwargs = {'disfluency': disfluency,
'conservative': conservative,
'disfluencies': set(['uh', 'um'])}
async = True
if b'async' in req.args and req.args[b'async'][0] == b'false':
async = False
# We need to make the transcription directory here, so that
# when we redirect the user we are sure that there's a place
# for them to go.
outdir = os.path.join(self.transcriber.data_dir, 'transcriptions', uid)
os.makedirs(outdir)
# Copy over the HTML
shutil.copy(get_resource('www/view_alignment.html'), os.path.join(outdir, 'index.html'))
result_promise = threads.deferToThreadPool(
reactor, reactor.getThreadPool(),
self.transcriber.transcribe,
uid, tran, audio, async, **kwargs)
if not async:
def write_result(result):
'''Write JSON to client on completion'''
req.setHeader("Content-Type", "application/json")
req.write(result.to_json(indent=2).encode())
req.finish()
result_promise.addCallback(write_result)
result_promise.addErrback(lambda _: None) # ignore errors
req.notifyFinish().addErrback(lambda _: result_promise.cancel())
return NOT_DONE_YET
req.setResponseCode(FOUND)
req.setHeader(b"Location", "/transcriptions/%s" % (uid))
return b''
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
def render_POST(self, req):
def quitGentle(restart):
req.setHeader("Content-Type", "application/json")
req.write("{ message: PROCESS_ENDED }")
req.finish()
time.sleep(0.5)
KILL_CMD_K3 = "taskkill /F /IM gentleK3.exe /T" if os.name == "nt" else "pkill -f gentleK3"
KILL_CMD_M3 = "taskkill /F /IM gentleM3.exe /T" if os.name == "nt" else "pkill -f gentleM3"
RESTART_CMD_WIN = [
"timeout", "1", "&&", "START", "/min", "cursesearch.exe"
]
RESTART_CMD_MAC = [
"sleep 1 && nohup ./cursesearch </dev/null >/dev/null 2>&1 &"
]
os.system(KILL_CMD_K3)
os.system(KILL_CMD_M3)
reactor.stop()
if restart is True:
if os.name == "nt":
subprocess.Popen(RESTART_CMD_WIN,
shell=True,
stdin=None,
stdout=None,
stderr=None,
close_fds=True,
cwd=os.path.dirname(sys.executable))
else:
subprocess.Popen(RESTART_CMD_MAC,
shell=True,
cwd=os.path.dirname(sys.executable))
sys.exit(0)
if "stop" in req.args and req.args["stop"][0]:
quitGentle(True)
if "stopNoRestart" in req.args and req.args["stopNoRestart"][0]:
quitGentle(False)
free_space = psutil.disk_usage(".").free / (1024 * 1024)
if free_space < 250:
shutil.rmtree(
os.path.join(self.transcriber.data_dir, 'transcriptions'))
uid = self.transcriber.next_id()
tran = req.args.get('transcript', [''])[0]
audio = req.args['audio'][0]
disfluency = True if 'disfluency' in req.args else False
conservative = True if 'conservative' in req.args else False
kwargs = {
'disfluency': disfluency,
'conservative': conservative,
'disfluencies': set(['uh', 'um'])
}
async = True
if 'async' in req.args and req.args['async'][0] == 'false':
async = False
# We need to make the transcription directory here, so that
# when we redirect the user we are sure that there's a place
# for them to go.
outdir = os.path.join(self.transcriber.data_dir, 'transcriptions', uid)
os.makedirs(outdir)
# Copy over the HTML
shutil.copy(get_resource('www/view_alignment.html'),
os.path.join(outdir, 'index.html'))
result_promise = threads.deferToThreadPool(reactor,
reactor.getThreadPool(),
self.transcriber.transcribe,
uid, tran, audio, async,
**kwargs)
if not async:
def write_result(result):
'''Write JSON to client on completion'''
req.setHeader("Content-Type", "application/json")
req.write(result.to_json(indent=2))
req.finish()
result_promise.addCallback(write_result)
result_promise.addErrback(lambda _: None) # ignore errors
req.notifyFinish().addErrback(lambda _: result_promise.cancel())
return NOT_DONE_YET
req.setResponseCode(FOUND)
req.setHeader(b"Location", "/transcriptions/%s" % (uid))
return ''
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
# Start a thread for the web server.
webthread = threading.Thread(target=serve.serve, args=(PORT, ))
webthread.start()
def open_browser():
webbrowser.open("http://localhost:%d/" % (PORT))
def open_about():
webbrowser.open("https://lowerquality.com/gentle")
dl = None
progress = None
fstpath = get_resource('exp/tdnn_7b_chain_online/graph_pp/HCLG.fst')
def download_full_language_model():
global progress, dl
fstdir = os.path.dirname(fstpath)
try:
os.makedirs(fstdir)
except OSError:
pass
progress = QtWidgets.QProgressBar()
progress.setRange(0, 100)
def render_POST(self, req):
uid = self.transcriber.next_id()
content = json.loads(req.content.read().decode("utf-8"))
tran = content["transcript"]
audioUrl = content["audioUrl"]
trim = content["trim"]
audioResult = requests.get(audioUrl, allow_redirects=True)
audio = audioResult.content
disfluency = True if b'disfluency' in req.args else False
conservative = True if b'conservative' in req.args else False
kwargs = {'disfluency': disfluency,
'conservative': conservative,
'disfluencies': set(['uh', 'um'])}
async_mode = True
if b'async' in req.args and req.args[b'async'][0] == b'false':
async_mode = False
# We need to make the transcription directory here, so that
# when we redirect the user we are sure that there's a place
# for them to go.
outdir = os.path.join(self.transcriber.data_dir, 'transcriptions', uid)
os.makedirs(outdir)
# Copy over the HTML
shutil.copy(get_resource('www/view_alignment.html'), os.path.join(outdir, 'index.html'))
result_promise = threads.deferToThreadPool(
reactor, reactor.getThreadPool(),
self.transcriber.transcribe,
uid, tran, audio, async_mode, trim, **kwargs)
if not async_mode:
def write_result(result):
'''Write JSON to client on completion'''
req.setHeader("Content-Type", "application/json")
req.setHeader("Access-Control-Allow-Origin", "*")
req.setHeader("Access-Control-Allow-Credentials", "True")
req.setHeader("Access-Control-Allow-Methods", "POST, GET, PUT, DELETE, OPTIONS")
req.write(result.to_json(indent=2).encode())
req.finish()
result_promise.addCallback(write_result)
result_promise.addErrback(lambda _: None) # ignore errors
req.notifyFinish().addErrback(lambda _: result_promise.cancel())
return NOT_DONE_YET
req.setHeader("Content-Type", "application/json")
req.setHeader("Access-Control-Allow-Origin", "*")
req.setHeader("Access-Control-Allow-Credentials", "True")
req.setHeader("Access-Control-Allow-Methods", "POST, GET, PUT, DELETE, OPTIONS")
result = { "uid": uid }
req.write(json.dumps(result).encode('utf-8'))
req.finish()
return NOT_DONE_YET
PORT = get_open_port(8765)
# Start a thread for the web server.
webthread = threading.Thread(target=serve.serve, args=(PORT,))
webthread.start()
def open_browser():
webbrowser.open("http://localhost:%d/" % (PORT))
def open_about():
webbrowser.open("https://lowerquality.com/gentle")
dl = None
progress = None
fstpath = get_resource('exp/tdnn_7b_chain_online/graph_pp/HCLG.fst')
def download_full_language_model():
global progress, dl
fstdir = os.path.dirname(fstpath)
try:
os.makedirs(fstdir)
except OSError:
pass
progress = QtWidgets.QProgressBar()
progress.setRange(0, 100)
trans.hide()
