def compute(self, x):
startstopsilence = es.StartStopSilence(threshold=-70)
for frame in es.FrameGenerator(x, frameSize=2048, hopSize=512,
startFromZero=True):
result = startstopsilence(esarray(frame))
solution = np.array([result[0], result[1]])
return solution * 512 / 44100.
def compute(self, x):
startstopsilence = es.StartStopSilence(threshold=-70)
for frame in es.FrameGenerator(x,
frameSize=2048,
hopSize=512,
startFromZero=True):
result = startstopsilence(esarray(frame))
solution = np.array([result[0], result[1]])
return solution * 512 / 44100.
def compute(self, *args):
x = args[1]
LPC = es.LPC(order=order, type='regular')
W = es.Windowing(size=frame_size, zeroPhase=False, type='triangular')
predicted = np.zeros(hop_size)
y = []
self.frames = []
self.errors = []
self.errors_filt = []
self.samples_peaking_frame = []
self.frame_idx = []
self.power = []
frame_counter = 0
for frame in es.FrameGenerator(x, frameSize=frame_size,
hopSize=hop_size,
startFromZero=True):
self.power.append(es.essentia.instantPower(frame))
self.frames.append(frame)
frame_un = np.array(frame[hop_size // 2: hop_size * 3 // 2])
frame = W(frame)
norm = np.max(np.abs(frame))
if not norm:
continue
frame /= norm
lpc_f, _ = LPC(esarray(frame))
lpc_f1 = lpc_f[1:][::-1]
for idx, i in enumerate(range(hop_size // 2, hop_size * 3 // 2)):
predicted[idx] = - np.sum(np.multiply(frame[i - order:i], lpc_f1))
error = np.abs(frame[hop_size // 2: hop_size * 3 // 2] - predicted)
threshold1 = times_thld * np.std(error)
med_filter = medfilt(error, kernel_size=kernel_size)
filtered = np.abs(med_filter - error)
mask = []
for i in range(0, len(error), sub_frame):
r = es.essentia.instantPower(frame_un[i:i + sub_frame]) > energy_thld
mask += [r] * sub_frame
mask = mask[:len(error)]
mask = np.array([mask]).astype(float)[0]
if sum(mask) == 0:
threshold2 = 1000 # just skip silent frames
else:
threshold2 = times_thld * (np.std(error[mask.astype(bool)]) +
np.median(error[mask.astype(bool)]))
threshold = np.max([threshold1, threshold2])
samples_peaking = np.sum(filtered >= threshold)
if samples_peaking >= 1:
y.append(frame_counter * hop_size / 44100.)
self.frame_idx.append(frame_counter)
self.frames.append(frame)
self.errors.append(error)
self.errors_filt.append(filtered)
self.samples_peaking_frame.append(samples_peaking)
frame_counter += 1
return np.array(y)
def compute(self, *args):
x = args[1]
LPC = es.LPC(order=order, type='regular')
W = es.Windowing(size=frame_size, zeroPhase=False, type='triangular')
predicted = np.zeros(hop_size)
y = []
self.frames = []
self.errors = []
self.errors_filt = []
self.samples_peaking_frame = []
self.frame_idx = []
self.power = []
frame_counter = 0
for frame in es.FrameGenerator(x,
frameSize=frame_size,
hopSize=hop_size,
startFromZero=True):
self.power.append(es.essentia.instantPower(frame))
self.frames.append(frame)
frame_un = np.array(frame[hop_size // 2:hop_size * 3 // 2])
frame = W(frame)
norm = np.max(np.abs(frame))
if not norm:
continue
frame /= norm
lpc_f, _ = LPC(esarray(frame))
lpc_f1 = lpc_f[1:][::-1]
for idx, i in enumerate(range(hop_size // 2, hop_size * 3 // 2)):
predicted[idx] = -np.sum(
np.multiply(frame[i - order:i], lpc_f1))
error = np.abs(frame[hop_size // 2:hop_size * 3 // 2] - predicted)
threshold1 = times_thld * np.std(error)
med_filter = medfilt(error, kernel_size=kernel_size)
filtered = np.abs(med_filter - error)
mask = []
for i in range(0, len(error), sub_frame):
r = es.essentia.instantPower(
frame_un[i:i + sub_frame]) > energy_thld
mask += [r] * sub_frame
mask = mask[:len(error)]
mask = np.array([mask]).astype(float)[0]
if sum(mask) == 0:
threshold2 = 1000 # just skip silent frames
else:
threshold2 = times_thld * (np.std(error[mask.astype(bool)]) +
np.median(error[mask.astype(bool)]))
threshold = np.max([threshold1, threshold2])
samples_peaking = np.sum(filtered >= threshold)
if samples_peaking >= 1:
y.append(frame_counter * hop_size / 44100.)
self.frame_idx.append(frame_counter)
self.frames.append(frame)
self.errors.append(error)
self.errors_filt.append(filtered)
self.samples_peaking_frame.append(samples_peaking)
frame_counter += 1
return np.array(y)
