def get_frame(self, file_name, frame_id):
"""Returns a frame from a specific param file."""
if self.last_file_name != file_name:
self.last_file_name = file_name
# print "FN", file_name
# find matching param file
param_file_name = self.get_param_file_name(file_name)
if param_file_name == None:
raise Exception("MLFMFCCOnlineAlignedArray: param_file_name cannot be None, file_name: " + file_name)
# print "PFN", param_file_name
# open the param file
try:
self.last_param_file_features = wave.open(param_file_name, 'r')
except AttributeError:
print "Error opening file:", param_file_name
if self.last_param_file_features.getnchannels() != 1:
raise Exception('Input wave is not in mono')
if self.last_param_file_features.getsampwidth() != 2:
raise Exception('Input wave is not in 16bit')
sample_rate = self.last_param_file_features.getframerate()
self.frame_size = int(sample_rate * self.windowsize / 10000000)
if self.frame_size > 1024:
self.frame_size = 2048
elif self.frame_size > 512:
self.frame_size = 1024
elif self.frame_size > 256:
self.frame_size = 512
elif self.frame_size > 128:
self.frame_size = 256
elif self.frame_size > 64:
self.frame_size = 128
self.frame_shift = int(sample_rate * self.targetrate / 10000000)
self.mfcc_front_end = MFCCFrontEnd(sample_rate, self.frame_size, usec0=self.usec0,
usedelta=self.usedelta, useacc=self.useacc,
n_last_frames=self.n_last_frames, mel_banks_only = self.mel_banks_only)
# print "FS", self.frame_size
self.last_param_file_features.setpos(max(frame_id * self.frame_shift - int(self.frame_size / 2), 0))
frame = self.last_param_file_features.readframes(self.frame_size)
# print "LN", len(frame)
frame = numpy.frombuffer(frame, dtype=numpy.int16)
try:
mfcc_params = self.mfcc_front_end.param(frame)
except ValueError:
print file_name, frame_id, len(frame)
raise
return mfcc_params
def get_frame(self, file_name, frame_id):
"""Returns a frame from a specific param file."""
if self.last_file_name != file_name:
self.last_file_name = file_name
# print "FN", file_name
# find matching param file
param_file_name = self.get_param_file_name(file_name)
# print "PFN", param_file_name
# open the param file
self.last_param_file_features = wave.open(param_file_name, 'r')
if self.last_param_file_features.getnchannels() != 1:
raise Exception('Input wave is not in mono')
if self.last_param_file_features.getsampwidth() != 2:
raise Exception('Input wave is not in 16bit')
sample_rate = self.last_param_file_features.getframerate()
self.frame_size = int(sample_rate * self.windowsize / 10000000)
if self.frame_size > 1024:
self.frame_size = 2048
elif self.frame_size > 512:
self.frame_size = 1024
elif self.frame_size > 256:
self.frame_size = 512
elif self.frame_size > 128:
self.frame_size = 256
elif self.frame_size > 64:
self.frame_size = 128
self.frame_shift = int(sample_rate * self.targetrate / 10000000)
self.mfcc_front_end = MFCCFrontEnd(
sample_rate,
self.frame_size,
usec0=self.usec0,
usedelta=self.usedelta,
useacc=self.useacc,
n_last_frames=self.n_last_frames)
# print "FS", self.frame_size
self.last_param_file_features.setpos(
max(frame_id * self.frame_shift - int(self.frame_size / 2), 0))
frame = self.last_param_file_features.readframes(self.frame_size)
# print "LN", len(frame)
frame = numpy.frombuffer(frame, dtype=numpy.int16)
mfcc_params = self.mfcc_front_end.param(frame)
return mfcc_params
def __init__(self, cfg):
self.cfg = cfg
self.audio_recorded_in = []
self.ffnn = TheanoFFNN()
self.ffnn.load(self.cfg['VAD']['ffnn']['model'])
self.log_probs_speech = deque(
maxlen=self.cfg['VAD']['ffnn']['filter_length'])
self.log_probs_sil = deque(
maxlen=self.cfg['VAD']['ffnn']['filter_length'])
self.last_decision = 0.0
if self.cfg['VAD']['ffnn']['frontend'] == 'MFCC':
self.front_end = MFCCFrontEnd(
self.cfg['Audio']['sample_rate'],
self.cfg['VAD']['ffnn']['framesize'],
self.cfg['VAD']['ffnn']['usehamming'],
self.cfg['VAD']['ffnn']['preemcoef'],
self.cfg['VAD']['ffnn']['numchans'],
self.cfg['VAD']['ffnn']['ceplifter'],
self.cfg['VAD']['ffnn']['numceps'],
self.cfg['VAD']['ffnn']['enormalise'],
self.cfg['VAD']['ffnn']['zmeansource'],
self.cfg['VAD']['ffnn']['usepower'],
self.cfg['VAD']['ffnn']['usec0'],
self.cfg['VAD']['ffnn']['usecmn'],
self.cfg['VAD']['ffnn']['usedelta'],
self.cfg['VAD']['ffnn']['useacc'],
self.cfg['VAD']['ffnn']['n_last_frames'] +
self.cfg['VAD']['ffnn']['n_prev_frames'],
self.cfg['VAD']['ffnn']['lofreq'],
self.cfg['VAD']['ffnn']['hifreq'],
self.cfg['VAD']['ffnn']['mel_banks_only'])
else:
raise ASRException('Unsupported frontend: %s' %
(self.cfg['VAD']['ffnn']['frontend'], ))
def __init__(self, model, filter_length, sample_rate, framesize,
frameshift, usehamming, preemcoef, numchans, ceplifter,
numceps, enormalise, zmeansource, usepower, usec0, usecmn,
usedelta, useacc, n_last_frames, n_prev_frames, lofreq,
hifreq, mel_banks_only):
self.audio_recorded_in = []
self.ffnn = TheanoFFNN()
self.ffnn.load(model)
self.log_probs_speech = deque(maxlen=filter_length)
self.log_probs_sil = deque(maxlen=filter_length)
self.last_decision = 0.0
self.front_end = MFCCFrontEnd(sample_rate, framesize, usehamming,
preemcoef, numchans, ceplifter, numceps,
enormalise, zmeansource, usepower, usec0,
usecmn, usedelta, useacc,
n_last_frames + n_prev_frames, lofreq,
hifreq, mel_banks_only)
self.framesize = framesize
self.frameshift = frameshift
