Frequency spectrum of sound using PyAudio, NumPy, and Matplotlib

README.rst

This Gist is about how I use PyAudio, NumPy, and Matplotlib to plot freqency spectrum of system sound or microphone.

You can read this blog post for more detail or watch this video:

merge.sh

#!/bin/bash

ffmpeg -i temp.mp4 -i temp.wav -vcodec copy -acodec libmp3lame sound-spectrum.mp4

sound-spectrum-wave.py

#!/usr/bin/env python
# Written by Yu-Jie Lin
# Public Domain
#
# Deps: PyAudio, NumPy, and Matplotlib
# Blog: http://blog.yjl.im/2012/11/frequency-spectrum-of-sound-using.html

from __future__ import print_function

import struct
import wave

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

TITLE = ''
WIDTH = 1280
HEIGHT = 720
FPS = 25.0

nFFT = 512
BUF_SIZE = 4 * nFFT
SAMPLE_SIZE = 2
CHANNELS = 2
RATE = 44100


def animate(i, line, wf, MAX_y):

  N = (int((i + 1) * RATE / FPS) - wf.tell()) / nFFT
  if not N:
    return line,
  N *= nFFT
  data = wf.readframes(N)
  print('{:5.1f}% - V: {:5,d} - A: {:10,d} / {:10,d}'.format(
    100.0 * wf.tell() / wf.getnframes(), i, wf.tell(), wf.getnframes()
  ))

  # Unpack data, LRLRLR...
  y = np.array(struct.unpack("%dh" % (len(data) / SAMPLE_SIZE), data)) / MAX_y
  y_L = y[::2]
  y_R = y[1::2]

  Y_L = np.fft.fft(y_L, nFFT)
  Y_R = np.fft.fft(y_R, nFFT)

  # Sewing FFT of two channels together, DC part uses right channel's
  Y = abs(np.hstack((Y_L[-nFFT / 2:-1], Y_R[:nFFT / 2])))

  line.set_ydata(Y)
  return line,


def init(line):

  # This data is a clear frame for animation
  line.set_ydata(np.zeros(nFFT - 1))
  return line,


def main():

  dpi = plt.rcParams['figure.dpi']
  plt.rcParams['savefig.dpi'] = dpi
  plt.rcParams["figure.figsize"] = (1.0 * WIDTH / dpi, 1.0 * HEIGHT / dpi)

  fig = plt.figure()

  # Frequency range
  x_f = 1.0 * np.arange(-nFFT / 2 + 1, nFFT / 2) / nFFT * RATE
  ax = fig.add_subplot(111, title=TITLE, xlim=(x_f[0], x_f[-1]),
                       ylim=(0, 2 * np.pi * nFFT ** 2 / RATE))
  ax.set_yscale('symlog', linthreshy=nFFT ** 0.5)

  line, = ax.plot(x_f, np.zeros(nFFT - 1))

  # Change x tick labels for left channel
  def change_xlabel(evt):
    labels = [label.get_text().replace(u'\u2212', '')
              for label in ax.get_xticklabels()]
    ax.set_xticklabels(labels)
    fig.canvas.mpl_disconnect(drawid)
  drawid = fig.canvas.mpl_connect('draw_event', change_xlabel)

  MAX_y = 2.0 ** (SAMPLE_SIZE * 8 - 1)
  wf = wave.open('temp.wav', 'rb')
  assert wf.getnchannels() == CHANNELS
  assert wf.getsampwidth() == SAMPLE_SIZE
  assert wf.getframerate() == RATE
  frames = wf.getnframes()

  ani = animation.FuncAnimation(
    fig, animate, int(frames / RATE * FPS),
    init_func=lambda: init(line), fargs=(line, wf, MAX_y),
    interval=1000.0 / FPS, blit=True
  )
  ani.save('temp.mp4', fps=FPS)

  wf.close()


if __name__ == '__main__':
  main()

sound-spectrum.py

#!/usr/bin/env python
# Written by Yu-Jie Lin
# Public Domain
#
# Deps: PyAudio, NumPy, and Matplotlib
# Blog: http://blog.yjl.im/2012/11/frequency-spectrum-of-sound-using.html

import struct
import wave

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

SAVE = 0.0
TITLE = ''
WIDTH = 1280
HEIGHT = 720
FPS = 25.0

nFFT = 512
BUF_SIZE = 4 * nFFT
FORMAT = pyaudio.paInt16
CHANNELS = 2
RATE = 44100


def animate(i, line, stream, wf, MAX_y):

  # Read n*nFFT frames from stream, n > 0
  N = max(stream.get_read_available() / nFFT, 1) * nFFT
  data = stream.read(N)
  if SAVE:
    wf.writeframes(data)

  # Unpack data, LRLRLR...
  y = np.array(struct.unpack("%dh" % (N * CHANNELS), data)) / MAX_y
  y_L = y[::2]
  y_R = y[1::2]

  Y_L = np.fft.fft(y_L, nFFT)
  Y_R = np.fft.fft(y_R, nFFT)

  # Sewing FFT of two channels together, DC part uses right channel's
  Y = abs(np.hstack((Y_L[-nFFT / 2:-1], Y_R[:nFFT / 2])))

  line.set_ydata(Y)
  return line,


def init(line):

  # This data is a clear frame for animation
  line.set_ydata(np.zeros(nFFT - 1))
  return line,


def main():

  dpi = plt.rcParams['figure.dpi']
  plt.rcParams['savefig.dpi'] = dpi
  plt.rcParams["figure.figsize"] = (1.0 * WIDTH / dpi, 1.0 * HEIGHT / dpi)

  fig = plt.figure()

  # Frequency range
  x_f = 1.0 * np.arange(-nFFT / 2 + 1, nFFT / 2) / nFFT * RATE
  ax = fig.add_subplot(111, title=TITLE, xlim=(x_f[0], x_f[-1]),
                       ylim=(0, 2 * np.pi * nFFT ** 2 / RATE))
  ax.set_yscale('symlog', linthreshy=nFFT ** 0.5)

  line, = ax.plot(x_f, np.zeros(nFFT - 1))

  # Change x tick labels for left channel
  def change_xlabel(evt):
    labels = [label.get_text().replace(u'\u2212', '')
              for label in ax.get_xticklabels()]
    ax.set_xticklabels(labels)
    fig.canvas.mpl_disconnect(drawid)
  drawid = fig.canvas.mpl_connect('draw_event', change_xlabel)

  p = pyaudio.PyAudio()
  # Used for normalizing signal. If use paFloat32, then it's already -1..1.
  # Because of saving wave, paInt16 will be easier.
  MAX_y = 2.0 ** (p.get_sample_size(FORMAT) * 8 - 1)

  frames = None
  wf = None
  if SAVE:
    frames = int(FPS * SAVE)
    wf = wave.open('temp.wav', 'wb')
    wf.setnchannels(CHANNELS)
    wf.setsampwidth(p.get_sample_size(FORMAT))
    wf.setframerate(RATE)

  stream = p.open(format=FORMAT,
                  channels=CHANNELS,
                  rate=RATE,
                  input=True,
                  frames_per_buffer=BUF_SIZE)

  ani = animation.FuncAnimation(
    fig, animate, frames,
    init_func=lambda: init(line), fargs=(line, stream, wf, MAX_y),
    interval=1000.0 / FPS, blit=True
  )

  if SAVE:
    ani.save('temp.mp4', fps=FPS)
  else:
    plt.show()

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

  if SAVE:
    wf.close()


if __name__ == '__main__':
  main()