批量将中文单个字音频转为图像（二维矩阵）

import numpy as np
import  librosa
import cv2
import matplotlib.pyplot as plt
import  os
import pypinyin
import torch
file_name_list=os.listdir("2500中文常用字发音")
label_class_list=sorted(set([ pypinyin.pinyin(one_name[:1])[0][0] for one_name in file_name_list]))
input_lable_list=[]
for one_name in file_name_list:
    print(one_name)

    audio,sr=librosa.load("D:/LSTM_poem/2500中文常用字发音/{}".format(one_name),16000)
    # x=np.linspace(6*np.pi+1,12*np.pi,200)
    # x_=np.linspace(1,6*np.pi+1,200)
    y=audio[1000:4000]
    total_res=np.hstack([y[::-1],y])
    mask=np.zeros([total_res.size,total_res.size])

    row,col=mask.shape
    m=row//2
    n=col//2
    mask[m,:]=total_res
    mask[m-1,:]=total_res
    mask[:,n]=total_res
    mask[:,n-1]=total_res
    x=0
    y=-1
    res=mask[m,0]
    mask[:,0]=[res]*(2*m-0*2)
    mask[:,-1]=[res]*(2*m-0*2)
    mask[0,:]=[res]*(2*m-0*2)
    mask[-1,:]=[res]*(2*m-0*2)
    for i in range(1,m-1):
        res = mask[m, i]
        mask[i:-i, i] = [res] * (2 * m - i * 2)
        mask[i:-i, -i-1] = [res] * (2 * m - i * 2)
        mask[i, i:-i] = [res] * (2 * m - i * 2)
        mask[-i-1, i:-i] = [res] * (2 * m - i * 2)

    input_lable_list.append([cv2.resize(mask,[128,128]),label_class_list.index(pypinyin.pinyin(one_name[:1])[0][0])])
torch.save({"data":input_lable_list,"voc":label_class_list},"train_audio_to_png.pth")
    # plt.imshow(cv2.resize(mask,[128,128]))
    # plt.show()
if __name__ == '__main__':
    pass