This repository contains basic code for a feed-forward linear layer written from scratch usig numpy and scipy for demonstration of the training of a simple speech recognition problem.

This contains all the source files including both the codes. Code.py is the file that contains the code for training the data. 250files.py contains the code for making 250 copies of each soundfile. To run the code, download the audio files from the folders in (https://drive.google.com/open?id=1hV5hcrKIM7oOEu3HkLIHvO7nDTFebDc_) and store it in a folder. In the same folder store 250files.py program. Run the program. The sample program shared is for reading the word "back". We can replace "back" by forward, left, right or stop and run the program for this. Once this is done, run the code.py program in the same folder. It will train the data and predict accuracy from test data.