Implemented in Python audio convolution, and make a simple HRTF effect
As a white users just getting started Python, writing this article is a waste me a lot of effort, but fortunately, patchwork everywhere on the Internet, I do give it to come out.
Ado, just start (entry to):
Audio convolution
My Python version is 2.7.13, install the required libraries:
pip install scipy
Import library:
from scipy.io import wavfile
from scipy.signal import fftconvolve
Serious operation begins:
file_path1 = '你的音频文件路径'
file_path2 = '你的第二个音频文件路径'
fs1, wav_data1 = wavfile.read(file_path1)
fs2, wav_data2 = wavfile.read(file_path2) #fs里保存的是头信息，wav_data保存的就是数据，它是以数组形式存在的
temp = fftconvolve(wav_data1,wav_data2) #将两个音频数据进行卷积
wavfile.write('temp要写入的文件路径'，44100，temp)
ps: If you encounter a warning: WavFileWarning:. Chunk (non-data) not understood, skipping it can clear the wav file metadata try
The effect of making HRTF
About this thing I will not be made, directly degree of your mother just fine
core requirement is as follows:
1, you have to determine the convolution hrir orientation point
2, it is best to double-channel file split into left and right two channels file performing convolution
HRIR data I use CIPIC, which is in the form of three-dimensional array data is saved in a file in the .mat.
Directly on the source code:
#coding=utf-8
#__Version__ 1.0
#Python 2.7.13
'''
Filename: fast_convolution.py
Project: hrtf制作
Author: o花间留步面馆o
FileCreated：Sunday, 22nd March 2020 14:32
Last Modified:Sunday, 23nd March 2020 1857
'''
import wave
import numpy as np
from scipy.io import wavfile
from scipy.io import loadmat
from scipy.signal import fftconvolve #所有操作需要的库
global filename_l
global filename_r
global filename_over_l
global filename_over_r #不想删了，怕又删出问题了。。。
print ("输入所有文件的同一个上级父目录：\n")
file_path = input()
print ("选择你的操作系统：\n 1,Mac/linux 2,windows")
tmp = input()
def ououou():
global filename_l
global filename_r
global filename_over_l
global filename_over_r
global hrir_l
global hrir_r
filename = file_path + '\\hrtf\\hrir_final.mat'
hrtfdata = loadmat(filename) #导入hrir数据
hrir_l = hrtfdata['hrir_l']
hrir_r = hrtfdata['hrir_r']
filename_l = file_path + '\\left.wav'
filename_r = file_path + '\\right.wav'
filename_over_l = file_path + '\\zuo.wav'
filename_over_r = file_path + '\\you.wav'
def lalala():
global filename_l
global filename_r
global filename_over_l
global filename_over_r
global hrir_l
global hrir_r
filename = file_path + '/hrtf/hrir_final.mat'
hrtfdata = loadmat(filename) #导入hrir数据
hrir_l = hrtfdata['hrir_l']
hrir_r = hrtfdata['hrir_r']
filename_l = file_path + '/left.wav'
filename_r = file_path + '/right.wav'
filename_over_l = file_path + '/zuo.wav'
filename_over_r = file_path + '/you.wav'
if tmp == 1:
lalala()
if tmp == 2:
ououou()
def project_1(n,c):
file_l = wave.open(filename_l,'rb')
file_r = wave.open(filename_r,'rb')
params_l = file_l.getparams()
params_r = file_r.getparams()
nchannels_l, sampwidth_l, framerate_l, nframes_l = params_l[:4]
nchannels_r, sampwidth_r, framerate_r, nframes_r = params_r[:4]
str_data_l = file_l.readframes(nframes_l) #读取byte数组
str_data_r = file_r.readframes(nframes_r)
file_l.close()
file_r.close()
wavdatal = np.fromstring(str_data_l, dtype = np.float32)
wavdatar = np.fromstring(str_data_r, dtype = np.float32) #byte数组转十进制数组
hrir_a = hrir_l[n,c,:]
hrir_b = hrir_r[n,c,:] #取方位点的数据
overl = fftconvolve(hrir_a,wavdatal)
overr = fftconvolve(hrir_b,wavdatar) #最重要的操作，卷积
wavfile.write(filename_over_l,44100,overl)
wavfile.write(filename_over_r,44100,overr) #写出文件
def project_3(n,c):
fs1, wavdatal = wavfile.read(filename_l)
fs2, wavdatar = wavfile.read(filename_r)
hrir_a = hrir_l[n,c, :]
hrir_b = hrir_r[n,c, :]
overl = fftconvolve(hrir_a,wavdatal)
overr = fftconvolve(hrir_b,wavdatar)
wavfile.write(filename_over_l,44100,overl)
wavfile.write(filename_over_r,44100,overr)
def project_2(n,c):
file_l = wave.open(filename_l,'rb')
file_r = wave.open(filename_r,'rb')
params_l = file_l.getparams()
params_r = file_r.getparams()
nchannels_l, sampwidth_l, framerate_l, nframes_l = params_l[:4]
nchannels_r, sampwidth_r, framerate_r, nframes_r = params_r[:4]
str_data_l = file_l.readframes(nframes_l)
str_data_r = file_r.readframes(nframes_r)
file_l.close()
file_r.close()
wavdatal = np.fromstring(str_data_l, dtype = np.short)
wavdatar = np.fromstring(str_data_r, dtype = np.short)
hrir_a = hrir_l[n,c,:]
hrir_b = hrir_r[n,c,:]
overl = fftconvolve(hrir_a,wavdatal)
overr = fftconvolve(hrir_b,wavdatar)
wavfile.write(filename_over_l,44100,overl)
wavfile.write(filename_over_r,44100,overr)
def project_4(n,a):
fs1, wavdatal = wavfile.read(filename_l)
fs2, wavdatar = wavfile.read(filename_r)
hrir_a = hrir_l[0,0,:]
hrir_b = hrir_r[0,0,:]
overl = fftconvolve(hrir_a,wavdatal)
overr = fftconvolve(hrir_b,wavdatar)
for n in range(1,20,a):
hrir_a = hrir_l[n,n,:]
hrir_b = hrir_r[n,n,:]
overl = fftconvolve(hrir_a,overl)
overr = fftconvolve(hrir_b,overr)
wavfile.write(filename_over_l,44100,overl)
wavfile.write(filename_over_r,44100,overr)
print("选择处理音频的方案：\n 1(成败参半，32位) 2（成败参半，16位） 3（稳定性ok） 4(实验性)")
o = 0
while o <= 3:
print("输入代表方案的数字:\n")
x = input()
if x <= 4:
break
o = o + 1
if x == 1:
n = int(raw_input("输入一个方位点：（0－24）").strip())
c = int(raw_input("再输一个：（0－24）").strip())
project_1(n,c)
if x == 2:
n = int(raw_input("输入一个方位点：（0－24）").strip())
c = int(raw_input("再输一个：（0－24）").strip())
project_2(n,c)
if x == 3:
n = int(raw_input("输入一个方位点：（0－24）").strip())
c = int(raw_input("再输一个：（0－24）").strip())
project_3(n,c)
if x == 4:
n = int(raw_input("输入一个方位点：（0－24）").strip())
a = int(raw_input("输每个方位点之间的间隔：（0－20）").strip())
project_4(n,a)
'''
注意事项：
1、要使用这个脚本，你需要将你的音频文件拆分成左，右两个单声道文件，并分别命名为left.wav和right.wav放在桌面上
2、在桌面上新建两个空文件，分别命名为zuo.wav和you.wav
3、你需要使用cipic的hrir数据，它的格式为.mat，将它放在桌面的\hrtf文件夹里并命名为hrir_final.mat
4、要运行这个脚本你还需要安装scipy库和numpy库，如果只使用方案3,4则只需要scipy库
python 2.x版本可以用 pip install scipy
python 3.x版本则是 pip3 install scipy #python3.x 我还没有测试
5、最后输出的数据需要限幅，我测试的结果它的输出最高到了792db
6、输入父目录时要按照如下格式：
windows: 'C:\\Adminmaster\\Desktop' ＃手里没有windows电脑，这样写也是网上看的，两个反斜杠转义成一个
Mac&linux: '/usr/share'
7、文件的所有输出都是44100hz,请尽量使用44100hz采样的音源
8、cipic的数据需要FQ下载，用wavfile读取的音频直接就以数组形式储存在name中
8、暂时就只想到这么多了，以后想到会补充的
'''
初学Python我也就只能写出这个水平的，肯定还有很多地方需要优化，还请大佬们不吝赐教
