def find_kiwi_regions(self, condition, segments, rate, min_no_ind_calls):
candidates = []
result = contiguous_regions(condition)
for start, end in result:
length = end - start
if length >= min_no_ind_calls:
region_start = segments[start][0]
region_end = segments[start + length - 1][1]
if self._density_above_threshold(region_start, region_end, length, rate):
candidates.append(Candidate(region_start, region_end))
else:
for i in np.arange(length - min_no_ind_calls):
region_start = i
region_end = i + min_no_ind_calls
if self._density_above_threshold(region_start, region_end, min_no_ind_calls, rate):
candidates.append(Candidate(region_start, region_end))
return candidates
def find_kiwi_regions(self, condition, segments, rate, min_no_ind_calls):
candidates = []
result = contiguous_regions(condition)
for start, end in result:
length = end - start
if length >= min_no_ind_calls:
region_start = segments[start][0]
region_end = segments[start + length - 1][1]
if self._density_above_threshold(region_start, region_end,
length, rate):
candidates.append(Candidate(region_start, region_end))
else:
for i in np.arange(length - min_no_ind_calls):
region_start = i
region_end = i + min_no_ind_calls
if self._density_above_threshold(
region_start, region_end, min_no_ind_calls,
rate):
candidates.append(
Candidate(region_start, region_end))
return candidates
def remove_clicks(signal, rate, window_size, margin):
"""
Clicks are bursts of energy. The fucntion will calculate signal energy over given window
size and eliminate regions of abnormaly high energy content.
Parameters
--------------
signal : 1d-array
Single-channel audio sample.
rate : int
Sample rate in Hz.
window_size : int
The number of data points used in each block for the energy calculation.
margin : float
How much (in seconds) the sample should be cut on both sides.
Returns
--------------
signal : 1d-array
Cleared signal.
"""
margin = margin * rate
overlap = window_size / 2.0
mask = np.ones(len(signal), dtype=bool)
if np.abs(signal.max()) > 2**14:
energy = calculate_energy(signal, window_size, overlap)
energy = sig.medfilt(energy, 15)
p = np.percentile(energy, 90)
condition = energy > 2*p
cont = contiguous_regions(condition)
for start, stop in cont:
start_idx = start * window_size - margin
stop_idx = stop * window_size + margin
mask[start_idx:stop_idx] = False
return signal[mask]
def remove_clicks(signal, rate, window_size, margin):
"""
Clicks are bursts of energy. The fucntion will calculate signal energy over given window
size and eliminate regions of abnormaly high energy content.
Parameters
--------------
signal : 1d-array
Single-channel audio sample.
rate : int
Sample rate in Hz.
window_size : int
The number of data points used in each block for the energy calculation.
margin : float
How much (in seconds) the sample should be cut on both sides.
Returns
--------------
signal : 1d-array
Cleared signal.
"""
margin = margin * rate
overlap = window_size / 2.0
mask = np.ones(len(signal), dtype=bool)
if np.abs(np.nanmax(signal)) > 2**14:
energy = calculate_energy(signal, window_size, overlap)
energy = sig.medfilt(energy, 15)
p = np.percentile(energy, 90)
condition = energy > 2*p
cont = contiguous_regions(condition)
for start, stop in cont:
start_idx = int(start * window_size - margin)
stop_idx = int(stop * window_size + margin)
mask[start_idx:stop_idx] = False
return signal[mask]