Blame view

README.md 1.05 KB
5dced4b57   Parcollet Titouan   Licence and readme
1
  Quaternion-Valued Convolutional Neural Networks for End-to-End Automatic Speech Recognition
f2d3bd141   Parcollet Titouan   Initial commit wi...
2
3
4
  =====================
  
  This repository contains code which reproduces experiments presented in
5dced4b57   Parcollet Titouan   Licence and readme
5
  the paper [link incoming]
f2d3bd141   Parcollet Titouan   Initial commit wi...
6
7
8
9
10
11
  
  Requirements
  ------------
  
  Install requirements for experiments with pip:
  ```
5dced4b57   Parcollet Titouan   Licence and readme
12
  pip install numpy tensorflow-gpu keras theano
f2d3bd141   Parcollet Titouan   Initial commit wi...
13
14
  ```
  Depending on your Python installation you might want to use anaconda or other tools.
5dced4b57   Parcollet Titouan   Licence and readme
15
  You can also go to the 'Installation' Section and execute the command for an automatic installation.
f2d3bd141   Parcollet Titouan   Initial commit wi...
16
17
18
  
  Installation
  ------------
5dced4b57   Parcollet Titouan   Licence and readme
19
  Install all the needed dependencies.
f2d3bd141   Parcollet Titouan   Initial commit wi...
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
  ```
  python setup.py install
  ```
  
  Experiments
  -----------
  
  1. Get help:
  
      ```
      python scripts/run.py train --help
      ```
  
  2. Run models:
  
      ```
5dced4b57   Parcollet Titouan   Licence and readme
36
      python scripts/run.py train --model {real,quaternion} --sf STARTFILTER --nl NUMBEROFLAYERS
f2d3bd141   Parcollet Titouan   Initial commit wi...
37
38
39
40
41
42
43
      ```
  
      Other arguments may be added as well; Refer to run.py train --help for
      
        - Optimizer settings
        - Dropout rate
        - Clipping
5dced4b57   Parcollet Titouan   Licence and readme
44
  	  - Saving prefix
f2d3bd141   Parcollet Titouan   Initial commit wi...
45
46
47
48
49
50
51
52
53
54
        - ...
  
  Citation
  --------
  
  Please cite our work as 
  
  ```
  
  ```