def process(self, line):
"""
Process a line from an SCP file, generate an MlfBlock Event from the
filename, and call the sendee with it
"""
# print line
fname = line.strip()
# We silently skip any blank or whitespace-only lines
if len(fname) == 0:
return
if self._prefix_subst is not None and fname.startswith(self._prefix_subst[0]):
fname = fname.replace(self._prefix_subst[0], self._prefix_subst[1], 1)
with open(fname, 'rb') as f:
(data, (kind, qualifiers), samp_period) = read_htk_audio_file(f)
self.send(HTKAudioBlockEvent(fname, (kind, qualifiers, samp_period), data))
def process(self, line):
"""
Process a line from an SCP file, generate an MlfBlock Event from the
filename, and call the sendee with it
"""
# print line
fname = line.strip()
# We silently skip any blank or whitespace-only lines
if len(fname) == 0:
return
if self._prefix_subst is not None and fname.startswith(
self._prefix_subst[0]):
fname = fname.replace(self._prefix_subst[0], self._prefix_subst[1],
1)
with open(fname, 'rb') as f:
(data, (kind, qualifiers), samp_period) = read_htk_audio_file(f)
self.send(
HTKAudioBlockEvent(fname, (kind, qualifiers, samp_period), data))
def score_feature_sequences_gmm(in_file_list, htk_model_file, out_dir):
'''
Read an HTK model file defining a set of GMMs and score the input feature files
based on these GMMs
'''
if not os.path.exists(out_dir):
os.makedirs(out_dir)
# Read ASCII mmf file
htk_mmf = open(htk_model_file, 'r')
models, hmm_mgr, gmm_mgr = htkmmf.read_htk_mmf_file(htk_mmf,
log_domain=True)
htk_mmf.close()
# Read feature files and score
in_list = open(in_file_list, 'r')
for fl in in_list:
fl = fl.rstrip('\r\n')
print fl
feature_file = open(fl, 'r')
features = htkaudio.read_htk_audio_file(feature_file)
feature_file.close()
scores = []
n_models = gmm_mgr.num_models
for md in range(n_models):
model_scores = gmm_mgr.get_model(md).score(features)
scores.append(model_scores)
score_file = os.path.join(
out_dir,
os.path.splitext(os.path.split(feature_file)[1])[0] + '.sco')
n_frames = len(scores[0])
sco_fl = open(score_file, 'w')
for count in range(n_frames):
for m_count in range(n_models):
sco_fl.write('{} '.format(scores[m_count][count]))
sco_fl.write('\n')
sco_fl.close()
in_list.close()
def score_feature_sequences_gmm(in_file_list, htk_model_file, out_dir):
'''
Read an HTK model file defining a set of GMMs and score the input feature files
based on these GMMs
'''
if not os.path.exists(out_dir):
os.makedirs(out_dir)
# Read ASCII mmf file
htk_mmf = open(htk_model_file,'r')
models, hmm_mgr, gmm_mgr = htkmmf.read_htk_mmf_file(htk_mmf, log_domain=True)
htk_mmf.close()
# Read feature files and score
in_list = open(in_file_list, 'r')
for fl in in_list:
fl = fl.rstrip('\r\n')
print fl
feature_file = open(fl,'r')
features = htkaudio.read_htk_audio_file(feature_file)
feature_file.close()
scores = []
n_models = gmm_mgr.num_models
for md in range(n_models):
model_scores = gmm_mgr.get_model(md).score(features)
scores.append(model_scores)
score_file = os.path.join(out_dir, os.path.splitext(os.path.split(feature_file)[1])[0]+'.sco')
n_frames = len(scores[0])
sco_fl = open(score_file,'w')
for count in range(n_frames):
for m_count in range(n_models):
sco_fl.write('{} '.format(scores[m_count][count]))
sco_fl.write('\n')
sco_fl.close()
in_list.close()
