Nonlinear Enhancement of Noisy Speech Using Dynamic Attractors in Neural Networks

Nonlinear Enhancement of Noisy Speech Using Dynamic Attractors in Neural Networks

Louiza Dehyadegari, Seyyed Ali Seyyedsalehi, Isar Nejadgholi

Abstract

In this paper speech denoising and phone recognition of enhanced signal is implemented using continues attractors and the capabilities of nonlinear recurrent neural networks in information retrieval. The recurrent neural network is first trained with clean speech and then is used to phone recognition of noisy speech with both stationary and nonstationary noise. In this work the efficiency of a nonlinear feedforward network is compared to the same one with a recurrent connection in hidden layer. The structure and training of this recurrent connection, is designed in such a way that the network learns to denoise the signal step by step, using properties of attractors it has formed, along with phone recognition. Using these connections, the recognition accuracy is improved 21% for stationary signal and 14% for nonstationary one with 0db SNR.

Keywords

Robust Recognition of Noisy Speech, Recurrent Neural Networks, Nonlinear Dynamics, Continues Attractors, Stationary Noise, Nonstationary Noise

References