def get_amplitude_envelope(data,
fs=30000.0,
lowpass=8000.0,
highpass=1000.0,
rectify_lowpass=600.0,
spec_sample_rate=200.0,
spec_freq_spacing=200.0,
mode="broadband"):
"""Compute an amplitude envelope of a signal
This can be done in two modes: "broadband" or "max_zscore"
Broadband mode is the normal amplitude envelope calcualtion. In broadband
mode, the signal is bandpass filtered, rectified, and then lowpass filtered.
Max Zscore mode is useful to detect calls which may be more localized in
the frequency domain from background (white) noise. In max_zscore mode, a
spectrogram is computed with spec_sample_rate time bins and
spec_freq_spacing frequency bins. For each time bin, the power in each
frequency bin is zscored and the max zscored value is assigned to the
envelope for that bin.
"""
spectral = True
filtered = highpass_filter(data.T, fs, highpass, filter_order=10).T
filtered = lowpass_filter(filtered.T, fs, lowpass, filter_order=10).T
if mode == "max_zscore":
t_spec, f_spec, spec, _ = spectrogram(
filtered,
fs,
spec_sample_rate=spec_sample_rate,
freq_spacing=spec_freq_spacing,
cmplx=False)
spec = np.abs(spec)
std = np.std(spec, axis=0)
zscored = (spec - np.mean(spec, axis=0)) / std
filtered = np.max(zscored, axis=0)
filtered -= np.min(filtered)
return scipy.signal.resample(filtered, len(data))
elif mode == "broadband":
# Rectify and lowpass filter
filtered = np.abs(filtered)
filtered = lowpass_filter(filtered.T, fs, rectify_lowpass).T
return filtered
else:
raise ValueError("Invalid amp_env_mode {}".format(amp_env_mode))
def compute(self):
result = {}
for ch in self.channels:
t_spec, f_spec, spec, _ = spectrogram(self.data[:, ch], *self.args, **self.kwargs)
result[ch] = (t_spec, f_spec, spec)
self.finished.emit(self.key, result)
# np.save("%sSeg2.npy" %fname[:-4], seg2)
# else:
# print('short seg')
# shutil.move('/Users/Alicia/Desktop/BirdCallProject/Freq_20_10000/filteredCalls/%s' %fname, \
# '/Users/Alicia/Desktop/BirdCallProject/Freq_20_10000/reSegmentedCalls/%s' %fname)
# if len(minimum) == 2:
# seg1 = sound[:minimum[0]]
# seg2 = sound[minimum[0]:minimum[1]]
# seg3 = sound[minimum[1]:]
# print(len(seg1), len(seg2), len(seg3))
# if len(seg1) > 1323 and len(seg2) > 1323 and len(seg3) > 1323:
# np.save("%sSeg1.npy" %fname[:-4], seg1)
# np.save("%sSeg2.npy" %fname[:-4], seg2)
# np.save("%sSeg3.npy" %fname[:-4], seg3)
# else:
# print('short seg')
# shutil.move('/Users/Alicia/Desktop/BirdCallProject/Freq_20_10000/filteredCalls/%s' %fname, \
# '/Users/Alicia/Desktop/BirdCallProject/Freq_20_10000/reSegmentedCalls/%s' %fname)
# plt.plot(sound_env/sound_env.max(), color="red", linewidth=2)
plt.figure()
(tDebug, freqDebug, specDebug, rms) = spectrogram(sound,
fs,
1000.0,
50,
min_freq=0,
max_freq=10000,
nstd=6,
cmplx=True)
plot_spectrogram(tDebug, freqDebug, specDebug)
plt.show()
buffer = 0.08
t_arr, sig = audio_signal.time_slice(
max(0, center_of_mass - buffer),
min(audio_signal.t_max, center_of_mass + buffer))
sig = sig - np.mean(sig, axis=0)
sig = bandpass_filter(sig.T, audio_signal.sampling_rate, 1000, 8000).T
specs = []
all_calls.append(sig)
for ch in range(sig.shape[1]):
# Sligtly lower resolution on the spectrograms can make this go faster
# Can increase the params to 1000, 50 for a higher resolution spectrogram
_, _, spec, _ = spectrogram(sig[:, ch],
audio_signal.sampling_rate,
500,
100,
min_freq=1000,
max_freq=8000,
cmplx=False)
specs.append(spec)
all_call_spectrograms.append(np.array(specs))
all_call_spectrograms = np.array(all_call_spectrograms)
all_calls = np.array(all_calls)
np.save(os.path.join(output_folder, "spectrograms.npy"),
all_call_spectrograms)
np.save(os.path.join(output_folder, "calls.npy"), all_calls)
print("""
import scipy.spatial as ss
import scipy.stats as sst
from scipy.special import digamma, gamma
from math import log, pi, exp
from scipy.integrate import odeint
from AttractorReconstructUtilities import TimeSeries3D, TimeDelayReconstruct
from matplotlib import style
from sciPlot import updateParams
from mpl_toolkits.axes_grid1 import make_axes_locatable
from matplotlib.ticker import MultipleLocator, FixedLocator
from soundsig.sound import WavFile, plot_spectrogram, spectrogram
updateParams()
X = np.load(
'/Users/Alicia/Desktop/BirdCallProject/Freq_250_12000/filteredCalls/LblBla4548_130418-DC-46.npy'
)
X = X[2900:4000]
(tDebug, freqDebug, specDebug, rms) = spectrogram(X,
44100,
2000.0,
50,
min_freq=0,
max_freq=22050,
nstd=6,
cmplx=True)
fig, ax = plt.subplots(figsize=(5, 2))
plot_spectrogram(tDebug, freqDebug, specDebug, ax=ax, colorbar=False)
ax.set_xlabel('t (s)')
plt.show()
#fig.savefig('Fig5Bifurcations.svg', dpi=300, bbox_inches='tight', transparent=True)
import numpy as np
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
from ipywidgets import interactive
from ipywidgets import FloatProgress
from IPython.display import display
import scipy.spatial as ss
import scipy.stats as sst
from scipy.special import digamma,gamma
from math import log,pi,exp
from scipy.integrate import odeint
from AttractorReconstructUtilities import TimeSeries3D, TimeDelayReconstruct
from matplotlib import style
from sciPlot import updateParams
from mpl_toolkits.axes_grid1 import make_axes_locatable
from matplotlib.ticker import MultipleLocator, FixedLocator
from soundsig.sound import WavFile, plot_spectrogram, spectrogram
updateParams()
TenSongs = np.load('/Users/Alicia/Desktop/SongAttractor/Attractor3D/TenSongs.npy', allow_pickle=True)
TenSongs = TenSongs[()]
#Attractor = TimeDelayReconstruct(TenSongs['BDY'][37300:48000], 2, 3)
(tDebug ,freqDebug ,specDebug , rms) = spectrogram(TenSongs['BDY'][37300:48000], 44100, 2000.0, 50, min_freq=0, max_freq=22050, nstd=6, cmplx=True)
fig, ax =plt.subplots(figsize=(5,2))
plot_spectrogram(tDebug, freqDebug, specDebug, ax=ax, colorbar=False)
ax.set_xlabel('t (s)')
plt.show()
#fig.savefig('Fig3BDYSpectrogram.svg', dpi=300, bbox_inches='tight', transparent=True)
def extract_spectrograms(
audio_signal,
intervals,
buffer=0.02, # buffer around original interval to keep
spec_buffer=0.04, # buffer so that everything thas the right padding.
):
"""Extract spectrograms from audio_signal denoted by intervals list
intervals is a list of (t_start, t_end) tuples
"""
_time = time.time()
print("Extracting spectrograms from {} intervals".format(len(intervals)))
# all_calls = []
all_call_spectrograms = []
for idx, (t1, t2) in enumerate(intervals):
print("Working on {}/{} ({:.2f}s elapsed)".format(
idx + 1, len(intervals),
time.time() - _time),
end="\r")
# Recentered signal with a small buffer of 40ms on either side
t_arr, sig = audio_signal.time_slice(
max(0, t1 - buffer), min(audio_signal.t_max, t2 + buffer))
sig = sig - np.mean(sig, axis=0)
sig = bandpass_filter(sig.T, audio_signal.sampling_rate, 1000, 8000).T
amp_env = get_amplitude_envelope(sig,
fs=audio_signal.sampling_rate,
lowpass=8000,
highpass=1000)
# Compute the temporal center of mass of the signal
center_of_mass = t1 - buffer + np.sum(
(t_arr * np.sum(amp_env, axis=1))) / np.sum(amp_env)
# Recentered signal with a small buffer of 40ms on either side
t_arr, sig = audio_signal.time_slice(
max(0, center_of_mass - spec_buffer),
min(audio_signal.t_max, center_of_mass + spec_buffer))
sig = sig - np.mean(sig, axis=0)
sig = bandpass_filter(sig.T, audio_signal.sampling_rate, 1000, 8000).T
specs = []
# all_calls.append(sig)
for ch in range(sig.shape[1]):
# Sligtly lower resolution on the spectrograms can make this go faster
# Can increase the params to 1000, 50 for a higher resolution spectrogram
_, _, spec, _ = spectrogram(sig[:, ch],
audio_signal.sampling_rate,
500,
100,
min_freq=1000,
max_freq=8000,
cmplx=False)
specs.append(spec)
all_call_spectrograms.append(np.array(specs))
all_call_spectrograms = np.array(all_call_spectrograms)
# all_calls = np.array(all_calls)
return all_call_spectrograms
#os.chdir('/Users/Alicia/Desktop/BirdCallProject/Freq_20_10000/dataAnalysis')
allSegments = np.load('allSegment_Len500.npy')
birdName = np.load('birdName.npy')
birdName = np.repeat(birdName, 3)
allFund = np.load('fund_Len500.npy')[:, None]
allFundFiltered = allFund[~np.isnan(allFund)]
tdata = np.arange(500)
num = np.int(500 / allFundFiltered[0] * 669)
(resampledSeg, resampledtdata) = resample(allSegments[0], num=num, t=tdata)
#plt.plot(tdata, allSegments[0])
#plt.plot(resampledtdata, resampledSeg, '--')
(tDebug, freqDebug, specDebug, rms) = spectrogram(allSegments[0],
44100,
1000.0,
100,
min_freq=0,
max_freq=10000,
nstd=6,
cmplx=True)
plot_spectrogram(tDebug, freqDebug, specDebug)
(tDebug, freqDebug, specDebug, rms) = spectrogram(resampledSeg,
44100,
1000.0,
100,
min_freq=0,
max_freq=10000,
nstd=6,
cmplx=True)
plt.figure()
plot_spectrogram(tDebug, freqDebug, specDebug)
