thewintersun / tensorflowbook

tensorflow教程每个章节的源码

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Question about Chapter 6 asr-lstm-ctc

lydiaji opened this issue · comments

In the example of Chapter 6 asr-lstm-ctc, it only has one wav file and one txt file to train. Now I would like to use Tedlium dataset to train, how can I rewrite the program? If I set the batch size as 64, then how about the inputs (its shape and its content)? Thanks so much!

要使用Tedlium 的数据集,一次性读入一个batch_size为64的数据,需要改动的代码会比较大。
大致过程如下:

  1. 在读取数据的时候,根据batchsize的大小,处理多个wav文件, 进行mfcc转换得到特征向量,将特征向量组合成矩阵。
  2. 修改网络结构,在网络进行计算的时候可以多维矩阵的计算;

不过修改的量比较大,几句话也说不清楚,可以参考这个:
https://github.com/thewintersun/asrtrain