语音预处理（相同话语不同人的对比）

显示同一段语音，不同说话人对比下效果
波形图

import wave
import matplotlib.pyplot as plt
import numpy as np
from scipy.io import wavfile
"""读取双通道波形并绘制波形图"""
f = wavfile.read("./test1.wav")
nframes = len(f[1])  #获取采样点数
nchannels = 1  #  通道数=1
framerate = f[0] #  获取采样频率
wave_data = f[1]  #  获取音频数据
wave_data = np.reshape(wave_data, [nframes, nchannels])
# 最后通过采样点数和取样频率计算出每个取样的时间
time = np.arange(0, nframes) * (1.0 / framerate)

plt.figure()
plt.subplot(3, 1, 1)
plt.plot(time, wave_data[:, 0])
plt.xlabel("time (seconds)")
plt.ylabel("Amplitude")
plt.title("people_one")
plt.grid()  # 标尺

plt.subplot(3, 1, 3)
f = wavfile.read("./test2.wav")
nframes = len(f[1])  #获取采样点数
nchannels = 1  #  通道数=1
framerate = f[0] #  获取采样频率
wave_data = f[1]  #  获取音频数据
wave_data = np.reshape(wave_data, [nframes, nchannels])
time = np.arange(0, nframes) * (1.0 / framerate)
plt.plot(time, wave_data[:, 0], c="g")
plt.xlabel("time (seconds)")
plt.ylabel("Amplitude")

plt.title("people_two")
plt.grid()
plt.show()

从时域转化到频域
看看效果是什么

import numpy as np
from scipy.io import wavfile
import matplotlib.pyplot as plt

def fft(path):
    sampling_freq, audio = wavfile.read(path)  # 读取文件

    audio = audio / np.max(audio)  # 归一化，标准化

    # 应用傅里叶变换
    fft_signal = np.fft.fft(audio)

    fft_signal = abs(fft_signal)

    # 建立时间轴
    Freq = np.arange(0, len(fft_signal))
    return Freq,fft_signal

plt.figure()
plt.subplot(3, 1, 1)
test1_freq , test1_signal = fft("./test1.wav")
plt.plot(test1_freq, test1_signal, color='blue')
plt.xlabel('Freq (in kHz)')
plt.ylabel('Amplitude')
plt.title("people_one")
plt.grid()  # 标尺

plt.subplot(3, 1, 3)
f = wavfile.read("./test2.wav")
test2_freq , test2_signal = fft("./test2.wav")
plt.plot(test2_freq, test2_signal, color='g')
plt.xlabel('Freq (in kHz)')
plt.ylabel('Amplitude')
plt.title("people_two")
plt.grid()

plt.show()

生成语谱图

import numpy as np
from scipy.io import wavfile
import matplotlib.pyplot as plt
from python_speech_features import mfcc, logfbank
import wave
def test(path):
    f = wave.open(path, 'rb')
    params = f.getparams()
    nchannels, sampwidth, framerate, nframes = params[:4]
    strdata = f.readframes(nframes)
    wavedata = np.fromstring(strdata, dtype=np.int16)
    wavedata = wavedata * 1.0 / (max(abs(wavedata)))
    wavedata = np.reshape(wavedata, [nframes, nchannels]).T
    f.close()
    return wavedata,framerate

plt.figure()
plt.subplot(3, 1, 1)
wave_1, framerate_1 = test('test1.wav')
plt.specgram(wave_1[0], Fs=framerate_1, scale_by_freq=True, sides = 'default')
plt.ylabel('Frequency(HZ)')
plt.xlabel('Time(s)')
plt.grid()  # 标尺

plt.subplot(3, 1, 3)
wave_2, framerate_2 = test('test2.wav')
plt.specgram(wave_2[0], Fs=framerate_2, scale_by_freq=True, sides = 'default')
plt.ylabel('Frequency(HZ)')
plt.xlabel('Time(s)')
plt.grid()

plt.show()

提取mfcc特征图

from scipy.io import wavfile
import matplotlib.pyplot as plt
from python_speech_features import mfcc, logfbank

def test(path):
    # 读取输入音频文件
    sampling_freq, audio = wavfile.read(path)
    # 提取MFCC特征
    mfcc_features = mfcc(audio, sampling_freq)
    mfcc_features = mfcc_features.T
    return  mfcc_features

# plt.figure()
# plt.subplot(3, 1, 1)
mfcc_1 = test('test1.wav')
plt.matshow(mfcc_1)
plt.xlabel('MFCC_windows  '+ str(mfcc_1.shape[0]) + '        ' +'MFCC_features_length  ' + str(mfcc_1.shape[1]))
plt.grid()  # 标尺

# plt.subplot(3, 1, 3)
mfcc_2 = test('test2.wav')
plt.matshow(mfcc_2)
plt.xlabel('MFCC_windows  '+ str(mfcc_2.shape[0]) + '        ' +'MFCC_features_length  ' + str(mfcc_2.shape[1]))
plt.grid()  # 标尺

plt.show()

本文地址：https://blog.csdn.net/qq_40703471/article/details/109860777