def rec_wav_file(asr, cfg, output_dir, wav_path):
""" Recognise speech in wav file and profile speech recognition.
The decoding and ASR output extraction times are estimated.
Args:
cfg (dict): Alex configuration with setting for speech recognition
wav_path (str): Path to Wave file which is recognised
Returns:
Tuple of decodeded ASR hypothesis, time of decoding, time of hypothesis extraction
"""
pcm = load_wav(cfg, wav_path)
frame = Frame(pcm)
start = time.time()
asr.rec_in(frame)
rec_in_end = time.time()
res = asr.hyp_out()
hyp_out_end = time.time()
save_lattice(asr.get_last_lattice(), output_dir, wav_path)
asr.flush()
return res, rec_in_end - start, hyp_out_end - rec_in_end
def rec_wav_file(output_dir, wav_path):
""" Recognise speech in wav file and profile speech recognition.
The decoding and ASR output extraction times are estimated.
Args:
cfg (dict): Alex configuration with setting for speech recognition
wav_path (str): Path to Wave file which is recognised
Returns:
Tuple of decodeded ASR hypothesis, time of decoding, time of hypothesis extraction
"""
pcm = load_wav(cfg, wav_path)
frame = Frame(pcm)
start = time.time()
asr.rec_in(frame)
rec_in_end = time.time()
res = asr.hyp_out()
hyp_out_end = time.time()
try:
save_lattice(asr.get_last_lattice(), output_dir, wav_path)
except AttributeError:
pass
asr.flush()
return res, rec_in_end - start, hyp_out_end - rec_in_end
def rec_wav_file(asr, cfg, wav_path):
pcm = load_wav(cfg, wav_path)
frame = Frame(pcm)
start = time.time()
asr.rec_in(frame)
rec_in_end = time.time()
res = asr.hyp_out()
hyp_out_end = time.time()
asr.flush()
return res, rec_in_end - start, hyp_out_end - rec_in_end
def send_wav(self, filename, stream=None):
"""Send given wavfile to the dialogue system as if it was said throught
microphone."""
# load wav
wav = load_wav(self.cfg, filename)
wav = various.split_to_bins(
wav, 2 * self.cfg['Audio']['samples_per_frame'])
# frame by frame send it
for frame in wav:
if stream is not None:
stream.write(frame)
self.audio_record.send(Frame(frame))
# send some silence so that VAD recognizes end of recording
for _ in range(10):
self.audio_record.send(Frame(b"\x00\x00" * self.cfg['Audio']['samples_per_frame']))
def get_tts_wav(self, voice, text):
"""Runs flite from the command line and gets the synthesized audio.
Note that the returned audio is in the re-sampled PCM audio format.
"""
handle, wav_file_name = mkstemp('TmpSpeechFile.wav')
if voice not in ['awb', 'rms', 'slt', 'kal', 'awb_time', 'kal16']:
voice = 'awb'
try:
subprocess.call("flite -voice %s -t \"%s\" -o %s 2> /dev/null" %
(voice, text, wav_file_name), shell=True)
wav = audio.load_wav(self.cfg, wav_file_name)
except:
raise TTSException("No data synthesized.")
return wav
def get_tts_wav(self, voice, text):
"""Runs flite from the command line and gets the synthesized audio.
Note that the returned audio is in the re-sampled PCM audio format.
"""
handle, wav_file_name = mkstemp('TmpSpeechFile.wav')
if voice not in ['awb', 'rms', 'slt', 'kal', 'awb_time', 'kal16']:
voice = 'awb'
try:
subprocess.call("flite -voice %s -t \"%s\" -o %s 2> /dev/null" %
(voice, text, wav_file_name),
shell=True)
wav = audio.load_wav(self.cfg, wav_file_name)
except:
raise TTSException("No data synthesized.")
return wav
""")
parser.add_argument('-c', "--configs", nargs='+',
help='additional configuration files')
args = parser.parse_args()
cfg = Config.load_configs(args.configs)
session_logger = cfg['Logging']['session_logger']
system_logger = cfg['Logging']['system_logger']
#########################################################################
#########################################################################
system_logger.info("Test of the AudioIO component\n" + "=" * 120)
wav = audio.load_wav(cfg, './resources/test16k-mono.wav')
# split audio into frames
wav = various.split_to_bins(wav, 2 * cfg['Audio']['samples_per_frame'])
# remove the last frame
aio_commands, aio_child_commands = multiprocessing.Pipe() # used to send aio_commands
audio_record, child_audio_record = multiprocessing.Pipe() # I read from this connection recorded audio
audio_play, child_audio_play = multiprocessing.Pipe() # I write in audio to be played
close_event = multiprocessing.Event()
aio = AudioIO(cfg, aio_child_commands, child_audio_record, child_audio_play, close_event)
aio.start()
count = 0
max_count = 2500
def main(dirname, outfname, cfg, skip=0, ignore_list_file=None):
"""
Arguments:
dirname -- the directory to search for WAVs
outfname -- path towards the file to output to
cfg -- a configuration dictionary (of the Config class)
skip -- how many wavs to skip (default: 0)
ignore_list_file -- a file open for reading whose lines specify path
globs for logs that should be ignored, or None if no such file
should be used. The format of this file is described in
some alex/corpustools scripts.
"""
# Fetch relevant config arguments.
frame_size = cfg['corpustools']['get_jasr_confnets']['frame_size']
rt_ratio = cfg['corpustools']['get_jasr_confnets']['rt_ratio']
sleep_time = rt_ratio * frame_size / 32000.
wavs = sorted(get_wav_fnames(dirname, ignore_list_file), key=itemgetter(1))
jul = None
try:
with codecs.open(outfname, 'a+', encoding='UTF-8') as outfile:
for wav_fname, wav_id in wavs[skip:]:
# Load the wav.
mywav = load_wav(cfg, wav_fname)
# Start Julius.
if jul is None:
jul, grep, errfile = start_julius(cfg, on_no_context)
# Insist on feeding all the input data to Julius, regardless of
# how many times it crashes.
exception = 1
while exception:
try:
for startidx in xrange(0, len(mywav), frame_size):
jul.rec_in(Frame(
mywav[startidx:startidx + frame_size]))
sleep(sleep_time)
# sleep(rt_ratio * len(mywav) / 32000.)
except socket.error as e:
# Julius crashing results in
# error: [Errno 104] Connection reset by peer
# Catch only that one.
if e.errno != 104:
raise e
exception = e
traceback.print_exc()
print "get_jasr_confnets: Restarting Julius."
clean_up(jul, grep, errfile)
jul, grep, errfile = start_julius(cfg, on_no_context)
else:
exception = None
exception = None
try:
hyp = jul.hyp_out()
except ASRException as e:
exception = e
except socket.error as e:
# Julius crashing results in
# error: [Errno 104] Connection reset by peer
# Catch only that one.
if e.errno != 104:
raise e
exception = e
if exception is not None:
traceback.print_exc()
clean_up(jul, grep, errfile)
jul = None
hyp = 'None'
exception = None
outfile.write('{id_} => {hyp!r}\n'.format(id_=wav_id, hyp=hyp))
sys.stderr.write('.')
sys.stderr.flush()
finally:
if jul is not None:
clean_up(jul, grep, errfile)
def rec_wav_file(self, wav_path):
pcm = load_wav(self.cfg, wav_path)
frame = Frame(pcm)
res = self.rec_wave(frame)
self.flush()
return res
'power_decision_frames': 25,
'power_decision_speech_threshold': 0.7,
'power_decision_non_speech_threshold': 0.2,
},
'Hub': {
'main_loop_sleep_time': 0.005,
},
'Logging': {
'output_dir': './tmp'
}
}
print "Test of the AudioIO and VAD components:"
print "=" * 120
wav = audio.load_wav(cfg, './resources/test16k-mono.wav')
# split audio into frames
wav = various.split_to_bins(wav, 2 * cfg['Audio']['samples_per_frame'])
# remove the last frame
aio_commands, aio_child_commands = multiprocessing.Pipe() # used to send commands to AudioIO
audio_record, child_audio_record = multiprocessing.Pipe() # I read from this connection recorded audio
audio_play, child_audio_play = multiprocessing.Pipe( ) # I write in audio to be played
vad_commands, vad_child_commands = multiprocessing.Pipe() # used to send commands to VAD
vad_audio_out, vad_child_audio_out = multiprocessing.Pipe()# used to read output audio from VAD
close_event = multiprocessing.Event()
aio = AudioIO(cfg, aio_child_commands, child_audio_record, child_audio_play, close_event)
vad = VAD(cfg, vad_child_commands, audio_record, vad_child_audio_out, close_event)
def rec_wav_file(self, wav_path):
pcm = load_wav(self.cfg, wav_path)
frame = Frame(pcm)
res = self.rec_wave(frame)
self.flush()
return res
def main(dirname, outfname, cfg, skip=0, ignore_list_file=None):
"""
Arguments:
dirname -- the directory to search for WAVs
outfname -- path towards the file to output to
cfg -- a configuration dictionary (of the Config class)
skip -- how many wavs to skip (default: 0)
ignore_list_file -- a file open for reading whose lines specify path
globs for logs that should be ignored, or None if no such file
should be used. The format of this file is described in
some alex/corpustools scripts.
"""
# Fetch relevant config arguments.
frame_size = cfg['corpustools']['get_jasr_confnets']['frame_size']
rt_ratio = cfg['corpustools']['get_jasr_confnets']['rt_ratio']
sleep_time = rt_ratio * frame_size / 32000.
wavs = sorted(get_wav_fnames(dirname, ignore_list_file), key=itemgetter(1))
jul = None
try:
with codecs.open(outfname, 'a+', encoding='UTF-8') as outfile:
for wav_fname, wav_id in wavs[skip:]:
# Load the wav.
mywav = load_wav(cfg, wav_fname)
# Start Julius.
if jul is None:
jul, grep, errfile = start_julius(cfg, on_no_context)
# Insist on feeding all the input data to Julius, regardless of
# how many times it crashes.
exception = 1
while exception:
try:
for startidx in xrange(0, len(mywav), frame_size):
jul.rec_in(
Frame(mywav[startidx:startidx + frame_size]))
sleep(sleep_time)
# sleep(rt_ratio * len(mywav) / 32000.)
except socket.error as e:
# Julius crashing results in
# error: [Errno 104] Connection reset by peer
# Catch only that one.
if e.errno != 104:
raise e
exception = e
traceback.print_exc()
print "get_jasr_confnets: Restarting Julius."
clean_up(jul, grep, errfile)
jul, grep, errfile = start_julius(cfg, on_no_context)
else:
exception = None
exception = None
try:
hyp = jul.hyp_out()
except ASRException as e:
exception = e
except socket.error as e:
# Julius crashing results in
# error: [Errno 104] Connection reset by peer
# Catch only that one.
if e.errno != 104:
raise e
exception = e
if exception is not None:
traceback.print_exc()
clean_up(jul, grep, errfile)
jul = None
hyp = 'None'
exception = None
outfile.write('{id_} => {hyp!r}\n'.format(id_=wav_id, hyp=hyp))
sys.stderr.write('.')
sys.stderr.flush()
finally:
if jul is not None:
clean_up(jul, grep, errfile)
