python3-声音处理

先来说下二进制读写文件，这需要struct库

#二进制文件读写
import struct
a=
b=-
# print(struct.pack("h",b))
# print(struct.pack("i",b))

f=open("bbb.bin","wb")
f.write(struct.pack("h",a))#对a装包，并写入
f.write(struct.pack("h",b))
f.close()

f=open("bbb.bin",'rb')
b1=f.read()
a=struct.unpack("h",b1)#解包操作
b2=f.read()
b=struct.unpack("h",b2)
print(a,b)

然后是通过python创造一段声音

import wave
import math
import struct
f=wave.open("v0.wav","w")
f.setframerate(8000)#声音频率
f.setnchannels(1)#声道数
f.setsampwidth(2)#声音宽度
t=0#时间
v=0.5#音量
dt=1/8000.0#录入声音的时间

while t<5:
    s=math.sin(t*math.pi*2*800)*v*32768#设置声音频率为-32768到32768，并且为正弦变化曲线
    s=int(s)
    fd=struct.pack("h",s)#二进制写入
    f.writeframes(fd)
    t+=dt
f.close()

运用这个原理可以写一个简单的声音文件，比如说两只老虎

import wave
import math
import struct
ff=wave.open("v1.wav","w")
ff.setframerate(8000)
ff.setnchannels(1)
ff.setsampwidth(2)

def wv(t=0,f=0,v=0.5,wf=ff,sr=8000):
    '''
    t:写入时长
    f:声音频率
    v：音量
    wf：一个可以写入的音频文件
    sr：采样率
    '''
    tt=0
    dt=1.0/sr
    while tt<=t:
        s=math.sin(tt*math.pi*2*f)*v*32768#采样，调节音量，映射到[-2^15,2^15)
        s=int(s)

        fd=struct.pack("h",s)#转换成8bit二进制数据
        wf.writeframes(fd)#写入音频文件
        tt+=dt#时间流逝

note={"":262,"":294,"":330,"":349,"":392,"":440,"":494,"6-":220,"":0}
n=[
    "","","","","","","","","",
    "","","","","","","","",
    "","","","","","","","","","","","","","",
    "","6-","","","","6-",""
]
tm=[
    2,2,2,2,2,2,2,2,1,
    2,2,2,1.5,2,2,2,2,
    1,1,1,1,2,2,1,1,1,1,1,1,2,1,
    2,2,2,2,2,2,2
]

for i in range(len(n)):
    wv(tm[i]/4.0,note[n[i]])

ff.close()

#河大校歌
import wave
import math
import struct
ff=wave.open("henu.wav","w")
ff.setframerate(8000)
ff.setnchannels(1)
ff.setsampwidth(2)

ff.getsampwidth

def wv(t=0,f=0,v=0.5,wf=ff,sr=8000):
    '''
    t:写入时长
    f:声音频率
    v：音量
    wf：一个可以写入的音频文件
    sr：采样率
    '''
    tt=0
    dt=1.0/sr
    while tt<=t:
        s=math.sin(tt*math.pi*2*f)*v*math.pow(math.e,-2.5*tt)*32768#采样，调节音量，映射到[-2^15,2^15)
        s=int(s)

        fd=struct.pack("h",s)#转换成8bit二进制数据
        wf.writeframes(fd)#写入音频文件
        tt+=dt#时间流逝

note={"-":0,"":0,"1---":49,"1#---":52,"2b---":52,"2---":55,"2#---":58,"3b---":58,"3---":62,"4---":65,"4#---":69,"5b---":69,"5---":73,"5#---":78,"6b---":78,"6---":82,"6#---":87,"7b---":87,"7---":92,"1--":98,"1#--":104,"2b--":104,"2--":110,"2#--":117,"3b--":117,"3--":123,"4--":131,"4#--":139,"5b--":139,"5--":147,"5#--":156,"6b--":156,"6--":165,"6#--":175,"7b--":175,"7--":185,"1-":196,"1#-":208,"2b-":208,"2-":220,"2#-":233,"3b-":233,"3-":247,"4-":262,"4#-":277,"5b-":277,"5-":294,"5#-":311,"6b-":311,"6-":330,"6#-":349,"7b-":349,"7-":370,"":392,"1#":415,"2b":415,"":440,"2#":466,"3b":466,"":494,"":523,"4#":554,"5b":554,"":587,"5#":622,"6b":622,"":659,"6#":698,"7b":698,"":740,"1+":784,"1#+":831,"2b+":831,"2+":880,"2#+":932,"3b+":932,"3+":988,"4+":1047,"4#+":1109,"5b+":1109,"5+":1175,"5#+":1245,"6b+":1245,"6+":1319,"6#+":1397,"7b+":1397,"7+":1480,"1++":1568,"1#++":1661,"2b++":1661,"2++":1760,"2#++":1865,"3b++":1865,"3++":1976,"4++":2093,"4#++":2217,"5b++":2217,"5++":2349,"5#++":2489,"6b++":2489}

n=[
    "5-","","","","7-",
    "","","","","",
    "","","","",
    "","","","-",
    "","","","7-","6-",
    "7-","","","","",
    "","","","",
    "","","","","-",
    "","","","",
    "","","","","",
    "","","","",
    "","","","7-","5-",
    "","","","","","-",
    "","","","-",
    "","","","","","-",
    "6-","7-","","-"
]
tm=[
    2,2,2,1,1,
    2,2,2,1,1,
    3,1,2,2,
    2,2,2,2,
    3,1,1,1,2,
    3,1,1,1,2,
    2,2,2,2,
    2,1,1,2,2,
    3,1,2,2,
    3,1,1,1,2,
    2,2,2,2,
    3,1,1,1,2,
    1,1,1,1,2,2,
    2,2,2,2,
    1,1,1,1,2,2,
    3,1,3,4
]

for i in range(len(n)):
    wv(tm[i]/2.0,note[n[i]])

ff.close()

接下来是比较高级的了，显示声音的音波

import matplotlib.pyplot as plt
import  wave
import struct

fig=plt.figure()
ax=fig.add_subplot(1,1,1)#如果参数是349的意思是：将画布分割成3行4列，图像画在从左到右从上到下的第9块
plt.ion()#使matplotlib的显示模式转换为交互（interactive）模式。即使在脚本中遇到plt.show()，代码还是会继续执行。
read_size=200#每次绘画的帧数
pass_f=0#跳过前pass的帧数
draw_channel=0
stay_time=1.0

file=wave.open("henu.wav","r")

frame_rate=file.getframerate()
frames = file.getnframes()#得到频率
channels=file.getnchannels()#得到声道数
draw_channel=draw_channel %channels
i=0
x_data=[x / read_size for x in range(read_size)]
fmt="h" * (read_size * channels)

while i<frames:
    fs=file.readframes(read_size)
    i+=read_size
    if pass_f > 0:
        pass_f-=1
        continue
    f_data = struct .unpack(fmt,fs)  # f_data中包含多个声道的声音
    y_data =[]
    for j in range(0,len(f_data)):
        if j%channels==draw_channel:
            y_data.append(f_data[j]/32768.0)
    lines = ax.plot(x_data, y_data, 'g-', lw=1)
    plt.pause(stay_time)
    ax.lines.remove(lines[0])
file.close()

这个是拟合一个图像的曲线，机器学习？？？？

#训练一个多项式，来拟合一条曲线
from PIL import Image
import matplotlib.pyplot as plt
import numpy as np
import random
import math
import tensorflow as tf

pix2=math.pi*2

img=np.array(Image.open("./p2.bmp"),dtype=np.int32)
#(249, 277, 3)图像的 高度，宽度，通道rgb
#Y中点125
data=[]
a=img[:,:,0]+img[:,:,1]+img[:,:,2]
h,w,_=np.shape(img)
nw=-1
for j in range(w):
    for i in range(h):
        if a[i,j]<200:
            data.append([j*1.0/w,(125.0-i)*2.0/w])
            if j%13==0 and nw!=j:
                nw=j
data_len=len(data)
#y=a1*sin(2pix*b1)+a2*sin(2pix*b2)+a3*sin(2pix*b3)...

#画图
fig=plt.figure()#生成图像框
ax=fig.add_subplot(1,1,1)#编号
ax.scatter([ w[0] for w in data],[w[1] for w in data],linewidths=1)
plt.ion()
#######################################################
#模型
w_size=8
batch_size=64

x=tf.placeholder(dtype=tf.float32, shape=[None,1])
y=tf.placeholder(dtype=tf.float32, shape=[None,1])
w=tf.Variable(tf.truncated_normal(shape=[w_size,1],mean=0.0,stddev=2/w_size,dtype=tf.float32))
#c=tf.constant([[(i+1)*2*math.pi for i in range(w_size)]],tf.float32)
c=tf.Variable(tf.constant([[(i+1)*2*math.pi for i in range(w_size)]],tf.float32))
res=tf.matmul(tf.sin(tf.matmul(x,c)),w)
loss=tf.reduce_mean(tf.abs((res-y)))

train_step=tf.train.AdamOptimizer(0.02).minimize(loss)
init=tf.global_variables_initializer()
sess=tf.Session()
sess.run(init)
for i in range(1,10000):
    rnddata=[random.randint(1,data_len-1) for w in range(batch_size)]
    x_d=[[data[p][0]] for p in rnddata]
    y_d=[[data[p][1]] for p in rnddata]
    ls,st=sess.run([loss,train_step],feed_dict={x:x_d,y:y_d})
    print(i,":",ls)

    if i%100==0:
        x_data=[[j/500.0] for j in range(500)]
        y_data,ww,cc=sess.run([res,w,c],feed_dict={x:x_data})
        print(ww)
        print(cc)
        lines=ax.plot([j[0] for j in x_data],[j[0] for j in y_data],'r-',lw=1)
        plt.pause(1)
        ax.lines.remove(lines[0])

最后是你输入文字然后让它发音，前提是你要有汉字表和发音表

t2pinyin={}
yunmu="āáǎàōóǒòēéěèīíǐìūúǔùǖǘǚǜ"
yindiao="123412341234123412341234"
rp="aaaaooooeeeeiiiiuuuuvvvv"
for i in range(1,10):#总共有十个文件，循环打开
    fn="./pinyin1/h"+str(i)+".txt"
    py=open(fn,"r",encoding="utf-8")#打开文件名  只读  utf-8
    line1=py.readline() #读取一行，再次调用就是下一行，先把第一行文字读进来
    line2=""    #读拼音
    while len(line2)<=10:#查找哪一行是拼音
        line2=py.readline()
    line1=line1.replace("\n","")#替换换行
    line2=line2.replace("\n","").replace("ü","v")
    py.close()
    pinyin=line2.split(" ")#变换成列表

    for j in range(len(line1)):
        word=line1[j]#word和pyj照应
        pyj=pinyin[j]
        flag=0
        for k in range(24):
            if pyj.find(yunmu[k])>=0:#查找拼音里的韵母
                pyj=pyj.replace(yunmu[k],rp[k])+yindiao[k]#替换，并标记声调
                flag=1
                break
        if flag==0:#没找到就是轻声
            pyj=pyj+"0"
        t2pinyin[word] = pyj
print(t2pinyin["大"],t2pinyin["家"],t2pinyin["好"])

s="我在学拍森"
import wave
wf=wave.open("py.wav","w")
wf.setframerate(16000)#设置频率的最大值
wf.setnchannels(2)#设置声道数
wf.setsampwidth(2)#设置声音宽度

for w in s:
    if w in t2pinyin:#找s中的每个字的读音
        vv=t2pinyin[w]
        file_name = "./pinyin/" + vv + ".wav"
        speak_file=wave.open(file_name,"r")
        data=speak_file.readframes(100000000)#将该文件下的前100000000个频率赋值给data
        wf.writeframes(data)
        speak_file.close()
wf.close()