使用keras完成语音识别

使用案例，训练两个类型的语音，然后测试，对CPU和内存要求不高。内存使用在 1G 左右

 

import wave
import matplotlib.pyplot as plt
import numpy as np
import os

import keras
from keras.models import Sequential
from keras.layers import Dense



# 加载数据集 和 标签[并返回标签集的处理结果]
def create_datasets():
    wavs=[]
    labels=[] # labels 和 testlabels 这里面存的值都是对应标签的下标，下标对应的名字在labsInd中
    testwavs=[]
    testlabels=[]

    labsInd=[]      ## 训练集标签的名字   0：seven   1：stop
    testlabsInd=[]  ## 测试集标签的名字   0：seven   1：stop

    path="E:/PycharmProjects/TensorFlow/实验/实验19-使用keras完成语音识别/wav/seven/"
    files = os.listdir(path)
    for i in files:
        # print(i)
        waveData = get_wav_mfcc(path+i)
        # print(waveData)
        wavs.append(waveData)
        if ("seven" in labsInd)==False:
            labsInd.append("seven")
        labels.append(labsInd.index("seven"))

    path="E:/PycharmProjects/TensorFlow/实验/实验19-使用keras完成语音识别/wav/stop/"
    files = os.listdir(path)
    for i in files:
        # print(i)
        waveData = get_wav_mfcc(path+i)
        wavs.append(waveData)
        if ("stop" in labsInd)==False:
            labsInd.append("stop")
        labels.append(labsInd.index("stop"))

    # 现在为了测试方便和快速直接写死，后面需要改成自动扫描文件夹和标签的形式
    path="E:/PycharmProjects/TensorFlow/实验/实验19-使用keras完成语音识别/wav/test1/"
    files = os.listdir(path)
    for i in files:
        # print(i)
        waveData = get_wav_mfcc(path+i)
        testwavs.append(waveData)
        if ("seven" in testlabsInd)==False:
            testlabsInd.append("seven")
        testlabels.append(testlabsInd.index("seven"))


    path="E:/PycharmProjects/TensorFlow/实验/实验19-使用keras完成语音识别/wav/test2/"
    files = os.listdir(path)
    for i in files:
        # print(i)
        waveData = get_wav_mfcc(path+i)
        testwavs.append(waveData)
        if ("stop" in testlabsInd)==False:
            testlabsInd.append("stop")
        testlabels.append(testlabsInd.index("stop"))

    wavs=np.array(wavs)
    labels=np.array(labels)
    testwavs=np.array(testwavs)
    testlabels=np.array(testlabels)
    return (wavs,labels),(testwavs,testlabels),(labsInd,testlabsInd)


def get_wav_mfcc(wav_path):
    f = wave.open(wav_path,'rb')
    params = f.getparams()
    # print("params:",params)
    nchannels, sampwidth, framerate, nframes = params[:4]
    strData = f.readframes(nframes)#读取音频，字符串格式
    waveData = np.fromstring(strData,dtype=np.int16)#将字符串转化为int
    waveData = waveData*1.0/(max(abs(waveData)))#wave幅值归一化
    waveData = np.reshape(waveData,[nframes,nchannels]).T
    f.close()

    # print(waveData)

    # plt.rcParams['savefig.dpi'] = 300 #图片像素
    # plt.rcParams['figure.dpi'] = 300 #分辨率
    # plt.specgram(waveData[0],Fs = framerate, scale_by_freq = True, sides = 'default')
    # plt.ylabel('Frequency(Hz)')
    # plt.xlabel('Time(s)')
    # plt.title('wa')
    # plt.show()

    ### 对音频数据进行长度大小的切割，保证每一个的长度都是一样的【因为训练文件全部是1秒钟长度，16000帧的，所以这里需要把每个语音文件的长度处理成一样的】
    data = list(np.array(waveData[0]))
    # print(len(data))
    while len(data)>16000:
        del data[len(waveData[0])-1]
        del data[0]
    # print(len(data))
    while len(data)<16000:
        data.append(0)
    # print(len(data))

    data=np.array(data)

    # 平方之后，开平方，取正数，值的范围在  0-1  之间
    data = data ** 2
    data = data ** 0.5

    return data


if __name__ == '__main__':
    (wavs,labels),(testwavs,testlabels),(labsInd,testlabsInd) = create_datasets()
    print(wavs.shape,"   ",labels.shape)
    print(testwavs.shape,"   ",testlabels.shape)
    print(labsInd,"  ",testlabsInd)

    # 标签转换为独热码
    labels = keras.utils.to_categorical(labels, 2)
    testlabels = keras.utils.to_categorical(testlabels, 2)
    print(labels[0]) ## 类似 [1. 0]
    print(testlabels[0]) ## 类似 [0. 0]

    print(wavs.shape,"   ",labels.shape)
    print(testwavs.shape,"   ",testlabels.shape)

    # 构建模型
    model = Sequential()
    model.add(Dense(512, activation='relu',input_shape=(16000,)))
    model.add(Dense(256, activation='relu'))
    model.add(Dense(64, activation='relu'))
    model.add(Dense(2, activation='softmax'))
    # [编译模型] 配置模型，损失函数采用交叉熵，优化采用Adadelta，将识别准确率作为模型评估
    model.compile(loss=keras.losses.categorical_crossentropy, optimizer=keras.optimizers.Adadelta(), metrics=['accuracy'])
    #  validation_data为验证集
    model.fit(wavs, labels, batch_size=124, epochs=5, verbose=1, validation_data=(testwavs, testlabels))

    # 开始评估模型效果 # verbose=0为不输出日志信息
    score = model.evaluate(testwavs, testlabels, verbose=0)
    print('Test loss:', score[0])
    print('Test accuracy:', score[1]) # 准确度

    model.save('asr_model_weights.h5') # 保存训练模型