Improved far-field speech recognition using Joint Variational Autoencoder