Simple Filter Design for Noise Removal

simple_filter.py

#!/usr/bin/env python

import scipy.fftpack as fftpack
import scipy.io.wavfile as wav
import numpy as np
import matplotlib.pylab as plt

# Reading wav file
AUDIO_FILE = '/home/abinash/hin_0044.wav'
input_file = open(AUDIO_FILE)
s_rate, wav_file = wav.read(input_file)

m, = wav_file.shape

# Plotting Time Domain Representation of Signal
figure_1 = plt.figure('Original Signal')
plt.subplot(2, 1, 1)
plt.title('Time Domain Signal')
plt.plot(np.arange(m)/float(s_rate), wav_file)
plt.xlabel('Time')
plt.ylabel('Amplitude')

# FFT of Signal
fft_signal = fftpack.fft(wav_file)
freq = fftpack.fftfreq(m)
# Plotting Frequency Domain Representation of Signal
plt.subplot(2, 1, 2)
plt.title('Frequency Domain')
plt.plot(freq*s_rate, fft_signal)
plt.xlabel('Frequency')
plt.ylabel('Spectrum')

# Noisy Signal
noisy_signal = wav_file + 100*np.random.randn(m)

figure_2 = plt.figure('Noisy Signal')
plt.subplot(2, 1, 1)
plt.title('Time Domain Noisy Signal')
plt.plot(np.arange(m)/float(s_rate), noisy_signal)
plt.xlabel('Time')
plt.ylabel('Amplitude')

fft_noisy_signal = fftpack.fft(noisy_signal)

plt.subplot(2, 1, 2)
plt.title('Frequency Domain Representation Noisy Signal')
plt.plot(freq*s_rate, fft_noisy_signal)
plt.xlabel('Frequency')
plt.ylabel('Spectrum')

filter_signal = [fft_noisy_signal[val] if freq[val]*s_rate < 3000 and
                 freq[val]*s_rate > -3000 else 0 for val in range(m)]

filter_signal_time = fftpack.ifft(filter_signal)

figure_fitler = plt.figure('Filtered Signal')
plt.plot(np.arange(m)/float(s_rate), filter_signal_time)

plt.show()