def _viterbiDecode(mfcc, bich_diar, thr_vit, wavFile, wdir, save_all):
wavName = wavFile[:-4]
vit_diar = viterbi.viterbi_decoding(mfcc, bich_diar, thr_vit)
if save_all:
vit_filename = os.path.join(wdir, wavName + '.d.seg')
Diar.write_seg(vit_filename, vit_diar)
return vit_diar
def create_seg_viterbi(self, cep, segment_dir):
#viterbi resegmentation
for file_name in os.listdir(segment_dir):
diar = Diar.read_seg(os.path.join(segment_dir, file_name))
vit_diar = viterbi.viterbi_decoding(cep, diar, self.vit_penalty)
Diar.write_seg(
os.path.join(self.results_vit_dir,
file_name + '.viterbi.{:.2f}'.format(-250)),
vit_diar)
def pyAudioDiar():
duration, result = aS.speaker_diarization(labelFileNameSound.get(),
int(labelNumberOfSpeakers.get()),
lda_dim=0,
plot_res=False)
show = 'diarizationExample'
input_show = labelFileNameSound.get()
input_sad = None
win_size = 250
thr_l = 2
thr_h = 3
thr_vit = -250
wdir = os.path.join('out', show)
if not os.path.exists(wdir):
os.makedirs(wdir)
fs = get_feature_server(input_show, feature_server_type='basic')
cep, _ = fs.load(show)
cep.shape
if input_sad is not None:
init_diar = Diar.read_seg(input_sad)
init_diar.pack(50)
else:
init_diar = segmentation.init_seg(cep, show)
seg_diar = segmentation.segmentation(cep, init_diar, win_size)
bicl_diar = segmentation.bic_linear(cep, seg_diar, thr_l, sr=False)
bic = hac_bic.HAC_BIC(cep, bicl_diar, thr_h, sr=False)
bich_diar = bic.perform(to_the_end=True)
vit_diar = viterbi.viterbi_decoding(cep, bich_diar, thr_vit)
resList = []
currentPosition = 0
for row in vit_diar:
speakerValue = int(row[1][1:])
while currentPosition < (row[3] + row[4]):
resList.append(speakerValue)
currentPosition += 20
currentPosition = 0
realityList = []
realityFile = pd.read_csv(labelFileNameSegment.get(),
delimiter='\t',
encoding='utf-8',
names=['start', 'end', 'speaker'])
for index, row in realityFile.iterrows():
speakerValue = int(row['speaker'][1:])
while currentPosition < row['end']:
realityList.append(int(speakerValue))
currentPosition += 0.2
plot.subplot(3, 1, 2)
plot.title("s4d:")
plot.plot(np.arange(0, duration, duration / len(resList)), resList, 'ro')
plot.subplot(3, 1, 1)
plot.title("Реальность:")
plot.plot(np.arange(0, duration, duration / len(realityList)), realityList,
'bo')
plot.subplot(3, 1, 3)
plot.title("pyPlotAudio:")
plot.plot(np.arange(0, duration, duration / len(result)), result, 'go')
plot.show()
def _viterbiDecode(mfcc, bich_diar, thr_vit):
return viterbi.viterbi_decoding(mfcc, bich_diar, thr_vit)