def extract_features(self, group):
assert group in ["enroll", "test"],\
"Invalid group name!! Choose either 'enroll', 'test'"
in_files = os.listdir(os.path.join(self.audio_dir, group))
feat_dir = os.path.join(self.feat_dir, group)
extractor = sidekit.FeaturesExtractor(
audio_filename_structure=os.path.join(self.audio_dir, group, "{}"),
feature_filename_structure=os.path.join(feat_dir, "{}.h5"),
lower_frequency=self.LOWER_FREQUENCY,
higher_frequency=self.HIGHER_FREQUENCY,
filter_bank=self.FILTER_BANK,
filter_bank_size=self.FILTER_BANK_SIZE,
window_size=self.WINDOW_SIZE,
shift=self.WINDOW_SHIFT,
ceps_number=self.CEPS_NUMBER,
vad=self.VAD,
snr=self.SNR_RATIO,
save_param=self.FEAUTRES,
keep_all_features=True)
show_list = np.unique(np.hstack([in_files]))
channel_list = np.zeros_like(show_list, dtype=int)
SKIPPED = []
for show, channel in zip(show_list, channel_list):
try:
extractor.save(show, channel)
except RuntimeError:
logging.info("SKIPPED")
SKIPPED.append(show)
continue
logging.info("Number of skipped files: " + str(len(SKIPPED)))
for show in SKIPPED:
logging.debug(show)
def make_feature_server(dirname, frame_size):
"""Return a Sidekit FeatureServer instance for this
experiement
"""
sampling_frequency = 16000
# window size must be twice the frame size to give the right number of FFT points but since
# we can't zero pad, we'll be taking in more of the signal in each frame
window_size = (2* frame_size+1) / sampling_frequency
shift = 0.008
# make a feature server to compute features over our audio files
extractor = sidekit.FeaturesExtractor(audio_filename_structure=dirname + "/{}.wav",
sampling_frequency=sampling_frequency,
lower_frequency=0,
higher_frequency=sampling_frequency/2,
filter_bank="lin",
filter_bank_size=frame_size,
window_size=window_size,
shift=shift,
ceps_number=20,
pre_emphasis=0.97,
save_param=["fb"],
keep_all_features=False)
server = sidekit.FeaturesServer(features_extractor=extractor,
sources=None,
dataset_list=["fb"],
keep_all_features=True)
return server
def extractFeatures(self, group):
"""
This function computes the acoustic parameters of audio files insied
"self.BASE_DIR/group":
for a list of audio files and save them to disk in a HDF5 format
"""
in_files = os.listdir(os.path.join(self.BASE_DIR, "audio", group))
feat_dir = os.path.join(self.BASE_DIR, "feat", group)
# Feature extraction
# lower_frequency: lower frequency (in Herz) of the filter bank
# higher_frequency: higher frequency of the filter bank
# filter_bank: type of fiter scale to use, can be "lin" or "log" (for linear of log-scale)
# filter_bank_size: number of filters banks
# window_size: size of the sliding window to process (in seconds)
# shift: time shift of the sliding window (in seconds)
# ceps_number: number of cepstral coefficients to extract
# snr: signal to noise ratio used for "snr" vad algorithm
# vad: Type of voice activity detection algorithm to use.
# It Can be "energy", "snr", "percentil" or "lbl".
# save_param: list of strings that indicate which parameters to save. The strings can be:
# for bottle-neck features and "vad" for the frame selection labels.
# keep_all_features: boolean, if True, all frames are writen; if False, keep only frames according to the vad label
extractor = sidekit.FeaturesExtractor(
audio_filename_structure=os.path.join(self.BASE_DIR, "audio",
group, "{}"),
feature_filename_structure=os.path.join(feat_dir, "{}.h5"),
lower_frequency=self.LOWER_FREQUENCY,
higher_frequency=self.HIGHER_FREQUENCY,
filter_bank=self.FILTER_BANK,
filter_bank_size=self.FILTER_BANK_SIZE,
window_size=self.WINDOW_SIZE,
shift=self.WINDOW_SHIFT,
ceps_number=self.CEPS_NUMBER,
vad=self.VAD,
snr=self.SNR_RATIO,
save_param=self.FEAUTRES,
keep_all_features=True)
# Prepare file lists
# show_list: list of IDs of the show to process
show_list = np.unique(np.hstack([in_files]))
# channel_list: list of channel indices corresponding to each show
channel_list = np.zeros_like(show_list, dtype=int)
# save the features in feat_dir where the resulting-files parameters
# are always concatenated in the following order:
# (energy, fb, cep, bnf, vad_label).
# SKIPPED: list to track faulty-files
SKIPPED = []
for show, channel in zip(show_list, channel_list):
try:
extractor.save(show, channel)
except RuntimeError:
logging.info("SKIPPED")
SKIPPED.append(show)
continue
logging.info("Number of skipped files: " + str(len(SKIPPED)))
for show in SKIPPED:
logging.debug(show)
def extractFeature(audio_dir, feature_dir, cep='mfcc', train=True):
'''
param:
audio_dir : path to audio dir
feature_dir : path to feature dir
cep : cepstrum type 'mfcc' or 'plp'
train : if True, extract feature from train set, else, from test set
return:
None
'''
# if feature dir not exists, create feature dir
if not os.path.exists(feature_dir):
os.mkdir(feature_dir)
wavlist = []
if train:
wav_list_file = os.getcwd() + '/log/aishell2_train.log'
else:
wav_list_file = os.getcwd() + '/log/aishell2_test.log'
if os.path.exists(wav_list_file):
with open(wav_list_file, 'r') as fobj:
for i in fobj:
wavlist.append(i[0:-1])
else:
wavlist = readDirStruct(audio_dir, train)
# prepare the necessary variables
showlist = np.asarray(wavlist)
channellist = np.zeros_like(showlist, dtype=int)
# create feature extractor
extractor = sidekit.FeaturesExtractor(
audio_filename_structure=audio_dir + '/{}.wav',
feature_filename_structure=feature_dir + "/{}.h5",
sampling_frequency=16000,
lower_frequency=133.3333,
higher_frequency=6955.4976,
filter_bank="log",
filter_bank_size=64,
window_size=0.025,
shift=0.01,
ceps_number=20,
vad="snr",
snr=40,
pre_emphasis=0.97,
save_param=["vad", "energy", "cep", "fb"],
feature_type=cep,
keep_all_features=True)
# save the feature
print('start extracting feature')
extractor.save_list(show_list=showlist,
channel_list=channellist,
num_thread=cpu_count() // 2)
print('extract feature done')
def make_feature_server():
"""Return a Sidekit FeatureServer instance for this
experiement
config: DATA_DIR, FEAT_DIR
"""
dd = config('DATA_DIR')
fd = config('FEAT_DIR')
# TODO: more of these settings should be derived from the config file
# make a feature server to compute features over our audio files
extractor = sidekit.FeaturesExtractor(audio_filename_structure=dd+'{}.wav',
feature_filename_structure=fd+"{}.h5",
sampling_frequency=None,
lower_frequency=200,
higher_frequency=3800,
filter_bank="log",
filter_bank_size=24,
window_size=0.025,
shift=0.01,
ceps_number=20,
vad="snr",
snr=40,
pre_emphasis=0.97,
save_param=["vad", "energy", "cep", "fb"],
keep_all_features=True)
server = sidekit.FeaturesServer(features_extractor=extractor,
feature_filename_structure=fd+"{}.h5",
sources=None,
dataset_list=["energy", "cep", "vad"],
mask="[0-12]",
feat_norm="cmvn",
global_cmvn=None,
dct_pca=False,
dct_pca_config=None,
sdc=False,
sdc_config=None,
delta=True,
double_delta=True,
delta_filter=None,
context=None,
traps_dct_nb=None,
rasta=True,
keep_all_features=True)
return server
def extract_speaker_data(self, speaker_files):
curr_speaker_num = 0
max_speakers = len(speaker_files.keys())
file_names = []
speaker_ids = []
fe = sidekit.FeaturesExtractor(
audio_filename_structure="{}.wav",
feature_filename_structure="common/data/training/i_vector/" +
self.speaker_list + "/feat/{}.h5",
sampling_frequency=None,
lower_frequency=200,
higher_frequency=3800,
filter_bank="log",
filter_bank_size=24,
window_size=0.025,
shift=0.01,
ceps_number=20,
vad="snr",
snr=40,
pre_emphasis=0.97,
save_param=["energy", "cep", "vad"],
keep_all_features=True)
# Crawl the base and all sub folders
for speaker in speaker_files.keys():
curr_speaker_num += 1
speaker_uid = curr_speaker_num
print('Extraction progress: %d/%d' %
(curr_speaker_num, max_speakers))
# Extract files
for full_path in speaker_files[speaker]:
file_name = os.path.splitext(os.path.basename(full_path))[0]
fe.save(show=speaker + '_' + file_name,
channel=0,
input_audio_filename=full_path,
output_feature_filename=None,
noise_file_name=None,
snr=10,
reverb_file_name=None,
reverb_level=-26.0)
file_names.append(speaker + '_' + file_name)
speaker_ids.append(str(speaker_uid))
return file_names, speaker_ids
def featureExtractor(subFolder_path):
e = sidekit.FeaturesExtractor(
audio_filename_structure=subFolder_path + '{}',
feature_filename_structure=None, #subFolder_path + '{}.h5',
sampling_frequency=8000,
lower_frequency=0,
higher_frequency=4000,
filter_bank="log",
filter_bank_size=24,
window_size=0.025,
shift=0.01,
ceps_number=20,
vad="snr",
snr=40,
pre_emphasis=0.97,
save_param=["vad", "energy", "cep", "fb"],
keep_all_features=False)
return e
def extract_feat(args):
# wav directory and feature directory
audio_folder = '/home/zeng/zeng/aishell/aishell2/ios/data/wav'
features_folder = '/home/zeng/zeng/aishell/aishell2/ios/data/feature'
wavlist = []
if os.path.exists(os.getcwd() + '/log/aishell2_wavlist.log'):
with open(os.getcwd() + '/log/aishell2_wavlist.log', 'r') as fobj:
for i in fobj:
wavlist.append(i[0:-1])
else:
wavlist = preprocess()
# prepare the necessary variables
showlist = np.asarray(wavlist)
channellist = np.zeros_like(showlist, dtype=int)
# create feature extractor
extractor = sidekit.FeaturesExtractor(
audio_filename_structure=audio_folder + '/{}.wav',
feature_filename_structure=features_folder + "/{}.h5",
sampling_frequency=16000,
lower_frequency=100.0,
higher_frequency=7000.0,
filter_bank="log",
filter_bank_size=64,
window_size=0.025,
shift=0.01,
ceps_number=24,
vad="snr",
snr=40,
pre_emphasis=0.97,
save_param=["vad", "energy", "cep", "fb"],
keep_all_features=True)
# save the feature
extractor.save_list(show_list=showlist,
channel_list=channellist,
num_thread=args.num_thread)
def make_mfcc_feats(self, wavList, input_file_list, output_feat_list, nj):
extractor = sidekit.FeaturesExtractor(audio_filename_structure=None,
feature_filename_structure=None,
sampling_frequency=8000,
lower_frequency=20,
higher_frequency=3700,
filter_bank='log',
filter_bank_size=24,
window_size=0.025,
shift=0.01,
ceps_number=20,
vad='percentil',
snr=40,
pre_emphasis=0.97,
save_param=["energy", "cep", "vad"],
keep_all_features=False
)
print("Start extracting the features")
channel_list = numpy.zeros(len(wavList), dtype='int')
extractor.save_list(show_list=wavList, channel_list=channel_list, audio_file_list=input_file_list,
feature_file_list=output_feat_list, num_thread=nj)
print("extracting the features success")
) as inputFile:
ubmList = inputFile.read().split('\n')
nameList = []
for a in ubmList:
nameList.append(a.rsplit('.')[0])
directory = '/home/adit/Desktop/DCASE2017-baseline-system-master/applications/data/TUT-acoustic-scenes-2017-development/audio/'
extractor = sidekit.FeaturesExtractor(
audio_filename_structure=
"/home/adit/Desktop/DCASE2017-baseline-system-master/applications/data/TUT-acoustic-scenes-2017-development/audio/{}.wav",
feature_filename_structure="/home/adit/Desktop/shikha_maam/Features/{}.h5",
sampling_frequency=16000,
lower_frequency=10,
higher_frequency=8000,
filter_bank="log",
filter_bank_size=40,
window_size=0.040,
ceps_number=12,
shift=0.020,
vad=None,
snr=None,
pre_emphasis=0.97,
save_param=["fb"],
keep_all_features=None)
show_list = np.unique([nameList])
channel_list = np.zeros_like(show_list, dtype=int)
logging.info("Extract features and save to disk")
extractor.save_list(show_list=show_list, channel_list=channel_list)
distribNb = 64 # number of Gaussian distributions for each GMM
audioDir = 'chunks_data/all_train/'
# Automatically set the number of parallel process to run.
# The number of threads to run is set equal to the number of cores available
# on the machine minus one or to 1 if the machine has a single core.
nbThread = max(multiprocessing.cpu_count()-1, 1)
extractor = sidekit.FeaturesExtractor(audio_filename_structure="chunks_data/all_train/{}",
feature_filename_structure="chunks_features/all_train/{}.h5",
sampling_frequency=None,
lower_frequency=200,
higher_frequency=4000,
filter_bank="log",
filter_bank_size=26,
window_size=0.025,
shift=0.01,
ceps_number=13,
# vad="snr",
# snr=10,
pre_emphasis=0.97,
save_param=["energy", "cep", "fb"],
keep_all_features=True)
server = sidekit.FeaturesServer(features_extractor=extractor,
feature_filename_structure="chunks_features/all_train/{}.h5",
# sources=None,
dataset_list=["energy", "cep", "fb"],
mask="[0-12]",
feat_norm="cmvn",
global_cmvn=None,
input_file_list.append(filepath)
output_feat_list.append(os.path.join(feat_fir, 'train', uttId + '.h5'))
output_feat_lines.append(
os.path.join(feat_fir, 'train', uttId + '.h5') + '\n')
with open(test_feats_scp_paath, 'w') as f:
f.writelines(output_feat_lines)
extractor_eval = sidekit.FeaturesExtractor(audio_filename_structure=None,
feature_filename_structure=None,
sampling_frequency=44100,
lower_frequency=20,
higher_frequency=3700,
filter_bank='log',
filter_bank_size=24,
window_size=0.025,
shift=0.01,
ceps_number=20,
vad='energy',
snr=40,
pre_emphasis=0.97,
save_param=["energy", "cep", "vad"],
keep_all_features=False)
print("Start extracting enroll features")
channel_list = np.zeros(len(enroll_list), dtype='int')
extractor_eval.save_list(show_list=enroll_list,
channel_list=channel_list,
audio_file_list=input_file_list,
feature_file_list=output_feat_list,
num_thread=4)
test_ndx = sidekit.Ndx('/info/home/larcher/task/3sess-pwd_eval_m_ndx.h5')
key = sidekit.Key('/info/home/larcher/task/3sess-pwd_eval_m_key.h5')
with open('/info/home/larcher/task/ubm_list.txt') as inputFile:
ubmList = inputFile.read().split('\n')
# Process the audio to save MFCC on disk
logging.info("Initialize FeaturesExtractor")
extractor = sidekit.FeaturesExtractor(
audio_filename_structure=audioDir + "/{}.wav",
feature_filename_structure="./features/{}.h5",
sampling_frequency=16000,
lower_frequency=133.3333,
higher_frequency=6955.4976,
filter_bank="log",
filter_bank_size=40,
window_size=0.025,
shift=0.01,
ceps_number=19,
vad="snr",
snr=40,
pre_emphasis=0.97,
save_param=["vad", "energy", "cep"],
keep_all_features=False)
# Get the complete list of features to extract
show_list = np.unique(
np.hstack([ubmList, enroll_idmap.rightids,
np.unique(test_ndx.segset)]))
channel_list = np.zeros_like(show_list, dtype=int)
logging.info("Extract features and save to disk")
def __init__(self,
input_filepath_structure=None,
feature_filename_structure=None,
mel_feature_filename_structure=None,
sampling_frequency=16000,
lower_frequency=200,
higher_frequency_mel=3800,
filter_bank="log",
filter_bank_size=24,
window_size=0.025,
shift=0.01,
ceps_number=20,
snr=40,
pre_emphasis=0.97):
self.__sk_lin_feature_extractor = sk.FeaturesExtractor(
audio_filename_structure=input_filepath_structure,
feature_filename_structure=feature_filename_structure,
lower_frequency=lower_frequency,
higher_frequency=sampling_frequency / 2,
filter_bank="lin",
filter_bank_size=filter_bank_size,
window_size=window_size,
shift=shift,
ceps_number=ceps_number,
vad="snr",
snr=snr,
save_param=["cep", "fb"],
sampling_frequency=sampling_frequency,
keep_all_features=True,
pre_emphasis=pre_emphasis,
)
self.__sk_mel_feature_extractor = sk.FeaturesExtractor(
audio_filename_structure=input_filepath_structure,
feature_filename_structure=mel_feature_filename_structure,
lower_frequency=lower_frequency,
higher_frequency=higher_frequency_mel,
filter_bank="log",
filter_bank_size=filter_bank_size,
window_size=window_size,
shift=shift,
ceps_number=ceps_number,
vad="snr",
snr=snr,
save_param=["vad", "energy", "cep", "fb"],
sampling_frequency=sampling_frequency,
keep_all_features=True,
pre_emphasis=pre_emphasis,
)
self.input_filepath_structure = input_filepath_structure
self.feature_filename_structure = feature_filename_structure
self.mel_feature_filename_structure = mel_feature_filename_structure
self.sampling_frequency = sampling_frequency
self.lower_frequency = lower_frequency
self.higher_frequency_mel = higher_frequency_mel
self.filter_bank_size = filter_bank_size
self.window_size = window_size
self.shift = shift
self.ceps_number = ceps_number
self.snr = snr
self.pre_emphasis = pre_emphasis
directory = os.fsencode("/home/adit/Desktop/DCASE2017-baseline-system-master/applications/data/TUT-acoustic-scenes-2017-development/audio")#"/home/adit/Desktop")# )
#os.chdir = "/home/adit/Desktop/DCASE2017-baseline-system-master/applications/data/TUT-acoustic-scenes-2017-development/audio/"
for file in os.listdir(directory):
filename = os.fsdecode(file)
if filename.endswith(".wav"):
# print(os.path.join(directory, filename))
extractor = sidekit.FeaturesExtractor(audio_filename_structure = "/home/adit/Desktop/DCASE2017-baseline-system-master/applications/data/TUT-acoustic-scenes-2017-development/audio/"+filename,
feature_filename_structure = "/home/adit/Desktop/DCASE2017-baseline-system-master/applications/data/TUT-acoustic-scenes-2017-development/audio_mfec_hdf5/"+filename+".h5",
sampling_frequency= 44100,
lower_frequency=20,
higher_frequency = 22000,
filter_bank = "log",
filter_bank_size = 40,
window_size = 0.040,
ceps_number = 12,
shift = 0.020,
vad = "snr",
snr = 40,
pre_emphasis= 0.97,
save_param = ["fb"],
keep_all_features = True)
extractor.save("")
file = h5py.File('/home/adit/Desktop/DCASE2017-baseline-system-master/applications/data/TUT-acoustic-scenes-2017-development/audio_mfec_hdf5/'+filename+'.h5','r+')
data = file['/fb']
#print(data)
tocsv = data[:,:]
data1 = file['/fb_mean']
tomean = data1[:,]
inputpath = "/home/aditya/Desktop/MFCCs"
show_list = []
channel_list = []
for filename in os.listdir(inputpath+"/accfiles"):
show_list += [os.path.splitext(filename)[0]]
channel_list += [0]
extractor = sidekit.FeaturesExtractor(audio_filename_structure=inputpath+"/accfiles/"+"{}.wav",
feature_filename_structure=inputpath+"/features/"+"{}.hd5",
sampling_frequency=8000,
lower_frequency=200,
higher_frequency=3800,
filter_bank="log",
filter_bank_size=24,
window_size=0.02,
shift=0.01,
ceps_number=20,
vad="snr",
snr=40,
pre_emphasis=0.97,
save_param=["energy", "cep"],
keep_all_features=True)
extractor.save_list(show_list=show_list,
channel_list=channel_list,
num_thread=6)
