This is for Digit Recognition using Microsoft DeepLearning Platform :CNTK
Training:
- This was trained to recognize 0-4 in English.
- to install CNTK on ubuntu use: https://github.com/Microsoft/CNTK/wiki/Setup-CNTK-on-Linux
- to install CNTK on Windows use: https://github.com/Microsoft/CNTK/wiki/Setup-CNTK-on-Windows
- features were extracted using "python_speech_features"with default values (https://github.com/jameslyons/python_speech_features)
- features were trained using Microsoft CNTK
- for getting ready the data for train:
- run loop.py- this file inputs the testwav.txt(line by line of training wav audios) and outputs all.txt containing MFCC features with labels at the beginning of each line.
- run uci2ctf.py with this command: python uci2ctf.py --input_file train.txt --features_start 1 --features_dim 533 --labels_start 0 --labels_dim 1 --num_labels 5 --output_file train2.txt
Test: Run text.bat on windows. This will execute: 1- rec.py to record a number 2- maintest.py to extract MFCC features(all.txt) 3- uci2ctf.py to change the type to "CNTKTextFormatReader" 4- cntk.exe configfile=..\Config\simple.cntk 5- mode.py to get the Mode of Predicted outputs. 6- print of the predicted digit to "output.txt"