def load_wav(wav_file):
# load audio file
pcm_data, samplerate = wav.read_wav(wav_file, 400)
sums, onset_times = onset_detection.detect_onset(pcm_data, samplerate,
False, True)
return pcm_data, onset_times, samplerate, sums
def load_wavs(self):
# load the raw training data files
trainingfiles = wav.get_training_filenames()
print (trainingfiles)
for pitch in trainingfiles:
pcm_data, self.samplerate = wav.read_wav(pitch, 400)
onsets = onset_detection.detect_onset(pcm_data, self.samplerate) # onsets in file
self.source_pcms.append(pcm_data)
self.onsetsall.append(onsets)
return [i for i in range(0, len(trainingfiles))]
x_training_path = sys.argv[2]
y_training_path = sys.argv[3]
x_validation_path = sys.argv[4]
y_validation_path = sys.argv[5]
generic_path = sys.argv[6]
use_cpu = True
if use_cpu:
os.environ['CUDA_VISIBLE_DEVICES'] = "-1"
profiler = Profiler()
x_training_wav = read_wav(x_training_path)
y_training_wav = read_wav(y_training_path)
x_validation_wav = read_wav(x_validation_path)
y_validation_wav = read_wav(y_validation_path)
profiler.stop('Done loading WAV files.')
if len(x_training_wav) != len(y_training_wav) or len(x_validation_wav) != len(
y_validation_wav):
raise Exception('Dry and wet WAVs must be same length.')
if mode == 'explore':
explorer = Explorer(x_training_wav, y_training_wav, x_validation_wav,
y_validation_wav, generic_path)
explorer.explore()
if len(sys.argv) != 6:
print('Usage:')
print(
f'python {sys.argv[0]} <dry training WAV file> <wet training WAV file> <dry validation WAV file> <wet validation WAV file> <output WAV file>'
)
sys.exit(1)
dry_training_wav_path = sys.argv[1]
wet_training_wav_path = sys.argv[2]
dry_validation_wav_path = sys.argv[3]
wet_validation_wav_path = sys.argv[4]
output_wav_path = sys.argv[5]
dry_training_wav = read_wav(dry_training_wav_path)
wet_training_wav = read_wav(wet_training_wav_path)
dry_validation_wav = read_wav(dry_validation_wav_path)
wet_validation_wav = read_wav(wet_validation_wav_path)
training_offset = 0
for i in range(0, len(dry_training_wav)):
if dry_training_wav[i] != 0:
training_offset = i
break
dry_training_wav = dry_training_wav[training_offset:]
wet_training_wav = wet_training_wav[training_offset:]
block_size = 128
offset = 21651
else:
clean_file_list = [clean_file_path]
noisy_file_list = [noisy_file_path]
cf.clear_csv_file(csv_file_path)
cf.write_csv_file(csv_file_path, "file name", "SNR,SSNR,PESQ")
for i in range(len(clean_file_list)):
if clean_path_isdir:
file_name = clean_file_list[i].replace(clean_file_path, "")
else:
file_name = os.path.basename(clean_file_path)
print("({}/{}){} /".format(i + 1, len(clean_file_list), file_name), end="")
# read train data file
clean_signal, clean_sample_rate = wav.read_wav(clean_file_list[i])
noisy_signal, noisy_sample_rate = wav.read_wav(noisy_file_list[i])
# different sample rate detect
if clean_sample_rate != noisy_sample_rate:
raise Exception(
"E: Different sample rate detected. clean({})/noisy({})".format(
clean_sample_rate, noisy_sample_rate))
SNR = cs.SNR(noisy_signal, clean_signal)
SSNR = cs.SSNR(noisy_signal, clean_signal, ssnr_frame_size)
PESQ = pesq(clean_signal, noisy_signal, clean_sample_rate)
print(" SNR:{} | SSNR:{} | PESQ:{}".format(SNR, SSNR, PESQ))
cf.write_csv_file(csv_file_path, file_name,
"{},{},{}".format(SNR, SSNR, PESQ))
raise Exception("ERROR: Target and source path is incorrect")
if target_path_isdir:
if not cf.compare_path_list(training_target_path, training_source_path, 'wav'):
raise Exception("ERROR: Target and source file list is not same")
training_target_file_list = cf.read_path_list(training_target_path, "wav")
training_source_file_list = cf.read_path_list(training_source_path, "wav")
else:
training_target_file_list = [training_target_path]
training_source_file_list = [training_source_path]
x_signal, y_signal = [], []
num_of_total_frame = 0
for i in range(len(training_target_file_list)):
# read train data file
target_signal, target_sample_rate = wav.read_wav(training_target_file_list[i])
source_signal, source_sample_rate = wav.read_wav(training_source_file_list[i])
target_signal = np.array(target_signal)
source_signal = np.array(source_signal)
size_of_target = target_signal.size
size_of_source = source_signal.size
# source & target file incorrect
if size_of_source != size_of_target:
raise Exception("ERROR: Input, output size mismatch")
if size_of_source < current_size:
raise Exception("ERROR: Input file length is too small")
if shift_size <= 0:
raise Exception("ERROR: Shift size is smaller or same with 0")
raise Exception("ERROR: Target and source path is incorrect")
if target_path_isdir:
if not cf.compare_path_list(training_target_path, training_source_path,
'wav'):
raise Exception("ERROR: Target and source file list is not same")
training_target_file_list = cf.read_path_list(training_target_path, "wav")
training_source_file_list = cf.read_path_list(training_source_path, "wav")
else:
training_target_file_list = [training_target_path]
training_source_file_list = [training_source_path]
x_signal, y_signal = [], []
num_of_total_frame = 0
for i in range(len(training_target_file_list)):
# read train data file
target_signal, target_sample_rate = wav.read_wav(
training_target_file_list[i])
source_signal, source_sample_rate = wav.read_wav(
training_source_file_list[i])
target_signal = np.array(target_signal)
source_signal = np.array(source_signal)
size_of_target = target_signal.size
size_of_source = source_signal.size
# source & target file incorrect
if size_of_source != size_of_target:
raise Exception("ERROR: Input, output size mismatch")
if size_of_source < current_size:
raise Exception("ERROR: Input file length is too small")
if shift_size <= 0:
raise Exception("ERROR: Shift size is smaller or same with 0")
return sign * normalized * MAX
def plot_onsets(sums, samplerate, onset_times):
# create a visual representation of the onsets
# debug function
print("plot onsets")
sums = np.asarray(sums)
time = np.arange(len(sums)) / samplerate
# draw a plot of the compressed audio samples
plt.figure(1)
plt.subplot(211)
plt.plot(time, sums)
# draw a vertical line at each detected onset
for i in onset_times:
xpos = i / samplerate
plt.plot([xpos, xpos], [0, 1], 'k-', lw=1)
plt.show()
if __name__ == '__main__':
if (len(sys.argv) > 1):
pcm_data, samplerate = wav.read_wav(sys.argv[1], 400)
detect_onset(pcm_data, samplerate, True)
def read_wav(f):
import wav
return wav.read_wav(files[f].file)
# make test data
if source_path_isdir:
test_source_file_list = cf.read_path_list(test_source_path, "wav")
if test_target_path_exist:
test_target_file_list = cf.read_path_list(test_target_path, "wav")
else:
test_target_file_list = test_source_file_list
else:
test_source_file_list = [test_source_path]
if test_target_path_exist:
test_target_file_list = [test_target_path]
else:
test_target_file_list = test_source_file_list
for i in range(len(test_source_file_list)):
test_source_signal, test_source_sample_rate = wav.read_wav(
test_source_file_list[i])
test_target_signal, test_target_sample_rate = wav.read_wav(
test_target_file_list[i])
test_target_signal = np.array(test_target_signal)
test_source_signal = np.array(test_source_signal)
test_size_of_source = test_source_signal.size
test_size_of_target = test_target_signal.size
if test_size_of_source != test_size_of_target:
raise Exception("ERROR: Test input, output size mismatch")
test_mod = (current_size -
(test_size_of_source % current_size)) % current_size
test_target_signal_padded = np.concatenate([
np.zeros(previous_size), test_target_signal,
np.zeros(future_size + test_mod)
def read_wav(f):
import wav
return wav.read_wav(files[f].file)
