Decompose a microphone input by frequency. Depends on PyAudio, Matplotib and NumPy

decompose.py

import pyaudio
import numpy as np
import matplotlib.pyplot as plt

CHUNKSIZE = 1024*2 # fixed chunk size

# initialize portaudio

p = pyaudio.PyAudio()
stream = p.open(format=pyaudio.paInt16, channels=1, rate=44100, input=True, frames_per_buffer=CHUNKSIZE)

# Wait to start recording

input("Press [Enter] to start recording")

result = np.array([], dtype=np.int16)

try:

    print("Press [Ctrl] + [C] to stop recording")

    while True:

        # Get the last two chunks into numpydata

        data = stream.read(CHUNKSIZE)
        newdata = np.fromstring(data, dtype=np.int16)
        result = np.append(result, newdata)

except KeyboardInterrupt:
    pass

print("Cleaning up...")

# Cleanup

stream.stop_stream()
stream.close()
p.terminate()

print("Parsing frequencies")

w = np.fft.fft(result)
freqs = np.abs((np.fft.fftfreq(len(w))*44100))

w = np.abs(w)

indices = np.argsort(w)[int(len(w)*0.99):]

bin_const = 10
indices = indices[len(indices)%bin_const:]

w = w[indices]
freqs = freqs[indices]

w = w[np.argsort(freqs)]
freqs = np.sort(freqs)

w = np.reshape(w, (-1, bin_const)).sum(axis=1)
freqs = np.average(np.reshape(freqs, (-1, bin_const)), axis=1)

w = w / np.sum(w)

freqdict = dict(zip(freqs, w))

import json
data = json.dumps(freqdict)

#print(data)

plt.plot(freqs, w, zorder=-1)

plt.plot(freqs, np.array([np.min(w)*2]*len(freqs)))

result = []

insidePeak = False

t = np.min(w)*3

row = []

for height, freq in zip(w, freqs):

    if freq < 100:
        continue

    if height > t:
        insidePeak = True
    else:
        if insidePeak:
            result.append(sorted(row)[-1])
            row = []
        insidePeak = False

    if insidePeak:
        row.append((height, freq)) 

#x = []
#y = []
#c = []

#for i, f in enumerate(np.convolve(w, [1, -0.5, -0.7, -1, 0, 1, 0.7, 0.5, -1])):
#    if f > 0:
#        try:
#            x.append(freqs[i-6])
#            y.append(w[i-6])
#            c.append(f*5000)
#        except:
#            pass

#peaks = sorted(list(zip(c, zip(x, y))))[-10:]

#x = [d[1][0] for d in peaks]
#y = [d[1][1] for d in peaks]
#c = [d[0] for d in peaks]



plt.scatter(np.array([x[1] for x in result]), np.array([x[0] for x in result]))

mul = 1/np.sum(np.array([x[0] for x in result]))

resarr = []

for x in result:
    resarr.append({"Freq":x[1], "Volume":mul*x[0]})
    #print("========================")
    #print("Frequency: "+str(x[1]))
    #print("Volume: "+str(mul*x[0]))
    #print("========================")

plt.show()

import sys
import json

if len(sys.argv[0]) > 1:
    with open(sys.argv[1], "w") as f:
        f.write(json.dumps(resarr))