Librosa - detect bat pulses in time domain

This is an early test to check if librosa can be used to detect bat pulses and sequences in time domain.

Bat sound is normally "sparse data" and music is "densed data". The idea here is to first check where there are sound peaks that differ from the background noise and after that use FFT for those parts only.

Hop-length is set to 1 ms in the examples below. Then it's easy to handle all used sampling frequencies from 192 to 500 kHz by dividing by 1000.

Alternatives in librosa that might work (but I have just started to use it and there are probably many more alternatives):

onset-detection. (71 ms)
rmse and localmax. (5 ms)
onset_strength and peak_pick. (68 ms)
peak_pick. (13 ms)

Note: When running "%%timeit" no values are stored that can be used in the following notebook cells.



In [ ]:



In [1]:

    
%matplotlib inline
import matplotlib.pyplot as plt
plt.rcParams['figure.figsize'] = (15, 4)



In [2]:

    
import numpy as np
import librosa

onset_detect



In [3]:

    
def test_onset_detect(y, hop_length):
    y2 = y.copy()
    rms_tot = np.sqrt(np.mean(y**2))
    y2[(np.abs(y) < (rms_tot * 1.5))] = 0.0
    onsets = librosa.onset.onset_detect(y=y2, sr=sr, hop_length=hop_length) 
    index_list = librosa.frames_to_samples(onsets, hop_length=hop_length)
    return index_list



In [4]:

    
y, sr = librosa.load('../data_in/Mdau_TE384.wav', sr=None)
index_list = test_onset_detect(y, 384)
print(len(index_list))
index_list









    



12






    Out[4]:





array([ 20736,  51072,  85632, 116736, 151680, 182784, 220416, 253056,
       283008, 310272, 342528, 373632])



In [5]:

    
%%timeit
index_list = test_onset_detect(y, 384)









    



52.2 ms ± 8.34 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)



In [6]:

    
y, sr = librosa.load('../data_in/Mdau_TE384.wav', sr=None)
index_list = test_onset_detect(y, 384)#Compare to original signal. Red dots are peaks.
plt.plot(y)
plt.scatter(index_list, [y[x:x+400].max() for x in index_list], color='r')
print(len(index_list))
plt.show()



In [7]:

    
y, sr = librosa.load('../data_in/Ppip_TE384.wav', sr=None)
index_list = test_onset_detect(y, 384)#Compare to original signal. Red dots are peaks.
plt.plot(y)
plt.scatter(index_list, [y[x:x+400].max() for x in index_list], color='r')
print(len(index_list))
plt.show()



In [8]:

    
y, sr = librosa.load('../data_in/Myotis-Plecotus-Eptesicus_TE384.wav', sr=None)
index_list = test_onset_detect(y, 384)#Compare to original signal. Red dots are peaks.
plt.plot(y)
plt.scatter(index_list, [y[x:x+400].max() for x in index_list], color='r')
print(len(index_list))
plt.show()



In [ ]:

rmse + localmax



In [9]:

    
def test_rmse_localmax(y, hop_length):
    y2 = y.copy()
    rms_tot = np.sqrt(np.mean(y**2))
    y2[(np.abs(y) < (rms_tot * 1.5))] = 0.0
    rmse = librosa.feature.rms(y=y2, hop_length=384, frame_length=1024, center=True)
    locmax = librosa.util.localmax(rmse.T)
    maxindexlist = []
    for index, a in enumerate(locmax):
        if a: maxindexlist.append(index)
    index_list = librosa.frames_to_samples(maxindexlist, hop_length=hop_length)
    return index_list



In [10]:

    
y, sr = librosa.load('../data_in/Mdau_TE384.wav', sr=None)
index_list = test_rmse_localmax(y, 384)
print(len(index_list))
index_list









    



13






    Out[10]:





array([ 21120,  51840,  85248,  86784, 117888, 152064, 183936, 220800,
       253824, 283392, 311040, 342912, 374400])



In [11]:

    
%%timeit
test_rmse_localmax(y, 384)









    



4.26 ms ± 444 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)



In [ ]:



In [12]:

    
y, sr = librosa.load('../data_in/Mdau_TE384.wav', sr=None)
index_list = test_rmse_localmax(y, 384) # Compare to original signal. Red dots are peaks.
plt.plot(y)
plt.scatter(index_list, [y[x:x+400].max() for x in index_list], color='r')
print(len(index_list))
plt.show()



In [13]:

    
y, sr = librosa.load('../data_in/Ppip_TE384.wav', sr=None)
index_list = test_rmse_localmax(y, 384) # Compare to original signal. Red dots are peaks.
plt.plot(y)
plt.scatter(index_list, [y[x:x+400].max() for x in index_list], color='r')
print(len(index_list))
plt.show()



In [14]:

    
y, sr = librosa.load('../data_in/Myotis-Plecotus-Eptesicus_TE384.wav', sr=None)
index_list = test_rmse_localmax(y, 384) # Compare to original signal. Red dots are peaks.
plt.plot(y)
plt.scatter(index_list, [y[x-200:x+200].max() for x in index_list], color='r')
print(len(index_list))
plt.show()



In [ ]:

onset_strength and peak_pick



In [ ]:



In [15]:

    
def test_onset_strength_and_peak_pick(y, hop_length):
    y2 = y.copy()
    rms_tot = np.sqrt(np.mean(y**2))
    y2[(np.abs(y) < (rms_tot * 1.5))] = 0.0
    onset_env = librosa.onset.onset_strength(y=y2, sr=sr,
                                             hop_length=384,
                                             aggregate=np.median)
    peak_index_list = librosa.util.peak_pick(onset_env, 3, 3, 3, 5, 0.5, 10)
    index_list = librosa.frames_to_samples(peak_index_list, hop_length=hop_length)
    return index_list



In [16]:

    
y, sr = librosa.load('../data_in/Mdau_TE384.wav', sr=None)
index_list = test_onset_strength_and_peak_pick(y, 384)
print(len(index_list))
index_list









    



12






    Out[16]:





array([ 21120,  51072,  86016, 116736, 152064, 183168, 220800, 253056,
       283008, 310656, 342528, 374016])



In [17]:

    
%%timeit
test_onset_strength_and_peak_pick(y, 384)









    



49.9 ms ± 5.13 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)



In [ ]:



In [18]:

    
y, sr = librosa.load('../data_in/Mdau_TE384.wav', sr=None)
index_list = test_onset_strength_and_peak_pick(y, 384) # Compare to original signal. Red dots are peaks.
plt.plot(y)
plt.scatter(index_list, [y[x:x+400].max() for x in index_list], color='r')
print(len(index_list))
plt.show()



In [19]:

    
y, sr = librosa.load('../data_in/Ppip_TE384.wav', sr=None)
index_list = test_onset_strength_and_peak_pick(y, 384) # Compare to original signal. Red dots are peaks.
plt.plot(y)
plt.scatter(index_list, [y[x:x+400].max() for x in index_list], color='r')
print(len(index_list))
plt.show()



In [20]:

    
y, sr = librosa.load('../data_in/Myotis-Plecotus-Eptesicus_TE384.wav', sr=None)
index_list = test_onset_strength_and_peak_pick(y, 384) # Compare to original signal. Red dots are peaks.
plt.plot(y)
plt.scatter(index_list, [y[x:x+400].max() for x in index_list], color='r')
print(len(index_list))
plt.show()



In [ ]:

peak_pick



In [21]:

    
def test_peak_pick(y, hop_length):
    y2 = y.copy()
    rms_tot = np.sqrt(np.mean(y**2))
    y2[(np.abs(y) < (rms_tot * 1.5))] = 0.0
    
    frames_per_ms = hop_length
    minmax_window = frames_per_ms / 4
    mean_window = frames_per_ms / 8
    sensitivity = rms_tot * 1.5 # 0.1
    skip_ms = 1

    index_list = librosa.util.peak_pick(y2, 
                                        minmax_window, minmax_window, 
                                        mean_window, mean_window, 
                                        sensitivity, 
                                        frames_per_ms * skip_ms)
    
    return index_list



In [22]:

    
y, sr = librosa.load('../data_in/Mdau_TE384.wav', sr=None)
index_list = test_peak_pick(y, 384)
print(len(index_list))
index_list









    



20






    Out[22]:





array([ 21140,  51114,  52002,  85686,  86705, 116787, 117211, 117771,
       151960, 182976, 184089, 220752, 253038, 254041, 283252, 284733,
       311067, 342957, 344337, 374260])



In [23]:

    
%%timeit
test_peak_pick(y, 384)









    



12.6 ms ± 520 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)



In [ ]:



In [24]:

    
y, sr = librosa.load('../data_in/Mdau_TE384.wav', sr=None)
index_list = test_peak_pick(y, 384) # Compare to original signal. Red dots are peaks.
plt.plot(y)
plt.scatter(index_list, [y[x:x+400].max() for x in index_list], color='r')
print(len(index_list))
plt.show()



In [25]:

    
y, sr = librosa.load('../data_in/Ppip_TE384.wav', sr=None)
index_list = test_peak_pick(y, 384) # Compare to original signal. Red dots are peaks.
plt.plot(y)
plt.scatter(index_list, [y[x:x+400].max() for x in index_list], color='r')
print(len(index_list))
plt.show()



In [26]:

    
y, sr = librosa.load('../data_in/Myotis-Plecotus-Eptesicus_TE384.wav', sr=None)
index_list = test_peak_pick(y, 384) # Compare to original signal. Red dots are peaks.
plt.plot(y)
plt.scatter(index_list, [y[x:x+400].max() for x in index_list], color='r')
print(len(index_list))
plt.show()



In [ ]:



In [ ]: