findpops.py

import shutil

from scipy import signal
from scipy.io import wavfile
import numpy as np 

debug_depop = True
if debug_depop:
    import matplotlib.pyplot as plt


def popfind(filename,channel=0):
    # read in file
    samplerate, data = wavfile.read(filename)
    print(samplerate)
    # collect data-type information and normalize
    original_type = data.dtype
    data_max = (np.iinfo(original_type).max)
    data = data.astype(float)
    data = data / data_max

    # compute basic properties
    length = data.shape[0] / samplerate
    num_channels = data.shape[1]
    time = np.linspace(0., length, data.shape[0])

    # stft of the channel
    # go through each band and, if its soft,
    # collect it to be averaged later
    f, t, Zxx = signal.stft(data[:, channel], samplerate, window=signal.get_window("hann",128),nperseg=128)
    zmean = np.zeros(Zxx[1,:].shape)
    znum = 0
    for i in range(Zxx.shape[0]):
    	z = 20*np.log10(np.abs(Zxx[i,:]))
    	z = z - np.mean(z[-500:])
        # find soft audio regions and use them for the average
        # arbitrary but works okay
    	if np.sum(z) < 10000:
    		zmean = np.add(zmean,z)
    		znum += 1

    # average all the soft bands
    if znum>0:
        zmean = zmean / znum 
    zmean = zmean - np.mean(zmean[-500:])
    z_std = np.std(zmean[-500:])

    peaks, _ = signal.find_peaks(zmean,height=z_std*6,prominence=10)
    if debug_depop:
        print(peaks)
        plt.plot(t,zmean)
        plt.plot(t[peaks],zmean[peaks],'o')
        plt.title("ch {}".format(channel))
        plt.show()

import sys
filename = sys.argv[1]
popfind(filename)