Fourier transform audio to add noise and remove noise (python two-dimensional fft2, ifft2)

Introduction

Standard Fast Fourier Transform

fft （a） [, n, axis, norm] ）	Computes the 1D discrete Fourier transform.
ifft （a） [, n, axis, norm] ）	Computes the 1D inverse discrete Fourier transform.
fft2 （a） [, s, axes, norm] ）	Computes the 2-D discrete Fourier transform.
ifft2 （a） [, s, axes, norm] ）	Computes the 2-D inverse discrete Fourier transform.
fftn （a） [, s, axes, norm] ）	Computes the n-dimensional discrete Fourier transform.
ifftn （a） [, s, axes, norm] ）	Computes the N-dimensional inverse discrete Fourier transform.

Real Fourier Transform

rfft （a） [, n, axis, norm] ）	Computes the 1D discrete Fourier transform of the actual input.
irfft （a） [, n, axis, norm] ）	Compute  rfft  .
rfft2 （a） [, s, axes, norm] ）	Computes the 2-D FFT of a real array.
irfft2 （a） [, s, axes, norm] ）	Compute  rfft2  .
rfftn （a） [, s, axes, norm] ）	Computes the N-dimensional discrete Fourier transform of the actual input.
irfftn （a） [, s, axes, norm] ）	Compute  rfftn  .

Hermitian Fast Fourier Transform

hfft （a） [, n, axis, norm] ）	Compute the FFT of a signal with Hermitian symmetry, i.e. the true spectrum.
ihfft （a） [, n, axis, norm] ）	Computes the inverse FFT of a signal with Hermitian symmetry.

helper routine

fftfreq n（n） [, d] ）	Returns the discrete Fourier transform sampling frequency.
rfftfreq n（n） [, d] ）	Returns the discrete Fourier transform sampling frequency (for rfft, irfft).
fftshift （x） [, axes] ）	Move the zero frequency component to the center of the spectrum.
ifftshift （x） [, axes] ）	Reverse  fftshift  .

 This article mainly adds noise to Xianxiafeng music and removes noise

Resource file name: spring.wav

Resource full path: https://github.com/fengjun321/noise.git

Core code: test3.py

If the resource file fails, you can go to Baidu to download other wav format music instead of spring.wav

# 用于保存音频
import wave
#数学库
import numpy as np
import matplotlib.pyplot as plt

#中文支持和布局调整
plt.rcParams['font.sans-serif']=['SimHei']
plt.rcParams['axes.unicode_minus']=False
plt.rcParams['figure.figsize']=(15,8)

plt.subplots_adjust(left=None, bottom=None, right=None, top=None,
                wspace=0.5, hspace=0.5)

def saveAudio(filename,data,params):
    with wave.open(filename + '.wav', 'wb') as wavfile:
        print(params)
        wavfile.setparams(params)
        wavfile.writeframes(bytes(data))

def wavread(path):
    wavfile = wave.open(path, "rb")
    params = wavfile.getparams()
    print(params)

    framesra, frameswav = params[2], params[3]
    datawav = wavfile.readframes(frameswav)
    wavfile.close()
    datause = np.frombuffer(datawav, dtype=np.short)
    datause.shape = -1, 2
    datause = datause.T
    time = np.arange(0, frameswav) * (1.0 / framesra)
    return datause, time, params

path = r"spring.wav"
wavdata, wavtime, params = wavread(path)

noise=np.random.rand(len(wavdata[0]))

noise_music = wavdata.copy() + noise
saveAudio("加噪后_spring", noise_music, params)

transformed=np.fft.fft2(noise_music)

avg1 = np.max(abs(transformed[0][1:]))/10000
avg2 = np.max(abs(transformed[1][1:]))/10000
transformed[0][np.where(abs(transformed[0])<=avg1)]=0+0j
transformed[1][np.where(abs(transformed[1])<=avg2)]=0+0j

noise_music = np.fft.ifft2(transformed).astype(int) #astype(int)很重要，过滤掉浮点过小信号

plt.subplot(231)
plt.title("原音频时序")
plt.plot(wavdata[0][4000:4500])

plt.subplot(232)
plt.title("原音频频域")
plt.plot(np.fft.fft(wavdata[0][4000:4200]))

plt.subplot(233)
plt.title("噪声音频时序")
plt.plot(noise_music[0][4000:4500])

plt.subplot(234)
plt.title("噪声音频频域")
plt.plot(np.fft.fft(noise_music[0][4000:4200]))

plt.show()
saveAudio("还原后_spring", noise_music, params)

Comparison of renderings

The downside is that there is no obvious noise in the picture, because the noise amplitude is small and it is not easy to observe. Readers can run the code directly to generate the corresponding audio file, and the sound effect playback is more obvious.