def main():
train_DB = read_DB_structure(c.TRAIN_DATAROOT_DIR)
transform = transforms.Compose([
truncatedinputfromMFB(),
totensor_DNN_input()
])
file_loader = read_MFB
speaker_list = sorted(set(train_DB['speaker_id']))
spk_to_idx = {spk: i for i, spk in enumerate(speaker_list)}
batch_size = 128
Dvector_train_dataset = Dvector_Dataset(DB=train_DB, loader=file_loader, transform=transform, spk_to_idx=spk_to_idx)
Dvector_train_loader = torch.utils.data.DataLoader(dataset=Dvector_train_dataset,
batch_size=batch_size,
shuffle=False)
from python_speech_features import fbank
import configure as c
from DB_wav_reader import read_DB_structure
import pickle
def normalize_frames(m, Scale=True):
if Scale:
return (m - np.mean(m, axis=0)) / (np.std(m, axis=0) + 2e-12)
else:
return (m - np.mean(m, axis=0))
# audiopath = c.TRAIN_WAV_DIR
audiopath = '/home/tuan/Documents/Train-Test-Data/public-test'
db = read_DB_structure(audiopath)
feat_and_label = {}
print(db["filename"])
for filename, label in zip(db["filename"], db["speaker_id"]):
print(filename)
print(label)
audio, sr = librosa.load(filename, sr=c.SAMPLE_RATE, mono=True)
filter_banks, energies = fbank(audio,
samplerate=c.SAMPLE_RATE,
nfilt=40,
winlen=0.025)
filter_banks = 20 * np.log10(np.maximum(filter_banks, 1e-5))
feature = normalize_frames(filter_banks, Scale=False)
feat_and_label['feat'] = feature