please install all libs required to this project

pip install numpy tensorflow scipy sounddevice random

To run, go to model/main.py change line 9 to your currenct directory

run main.py file

Note: sample_generator.py was used to record sound to fill our database.

This model was based on Tensorflow model, it will recognize one word each time. The audio file format is .wav in a 16000Hz. It was used 80% of data for trainning and 20% for validation. This model convert audio to spectrogram. Since spectrogram is an image, it is possible to use CNN(Convolutional Neuro Network) as a model. The model consists in two convolutional 2D CNN with ReLU activation, one layer for dimentionality reduction, one layer of dropout for overfiting reduction and the output model using the number of labels.