This repository consists of application of Deep Learning Models like DNN, CNN (1D and 2D), RNN (LSTM and GRU) and Variational Autoencoders written from scratch in tensorflow.
Folder information:
DNN_MNIST - Implementation of a Deep Neural Network (DNN) consisting of 4 layers along with data reduction and visualization using tSNE and PCA.
DNN_Speech_Denoising - Implementation of a Deep Neural Network (DNN) consisting of 4 layers with SNR value of 13.07 dB.
CNN_1D_Speech_Denoising - Implementation of a Convolutional Neural Network (CNN) using 1D features of the audio with SNR value of 16.28 dB.
CNN_2D_Speech_Denoising - Implementation of a Convolutional Neural Network (CNN) using 2D features of the audio with SNR value of 14.63 dB.
RNN_Speech_Denoising - Implementation of a Recurrent Neural Network (RNN) using a Long short-term memory (LSTM) model with SNR value of 11.19 dB.
SVD_Network_Compression - Implementation of a DNN along with Singular Value Decomposition on weight matrices for Network Compression.
Siamese_GRU_Speaker_Verification - Implementation of a Siamese Network using a GRU model for Speaker Verification.
VAE - Implementation of Variational Autoencoders to generate images and identify the effects to the 7's.