Speech protected active noise control system

Noise can be described as the unwanted sounds. This unwanted sound causes permanent hearing damage when it is too loud. Therefore, the usage of hearing equipment is needed in noisy environments. The usual noise Control systems eliminate both unwanted sounds and desired sounds. In this thesis, an active noise control system which preserves the speech while eliminating the noise in noisy environments is proposed. With the help of this system a worker will be able to hear speech in a noisy environment. The software of the proposed system which is the combination of Voice Activity Detection (VAD), Independent Component Analysis (ICA) and Active Noise Control (ANC) blocks is developed. As a first step, VAD block determines that the speech is contained in the environment or not. At the second step, ICA block separates the signals only if VAD block makes a decision that the speech exists there. From the output of the ICA block two signals are obtained. To decide which one is the noise, variance of autocorrelation which is a simple feature is used. Then the noise is eliminated by the ANC block. ANC block is designed with the Fx-LMS algorithm. The system performance is tested on a dataset which is generated by taking the several samples from different noise types. These sounds are mixed with the speech by using a matrix which is nonsingular symmetric at five different SNR levels. The dataset contains 11600 samples of which length is 12 seconds. Simulations show that the performance of the proposed system highly depends on the VAD block. If this block performs well, ANC protects the speech and eliminates the noise. To obtain better performance a new feature based on Mel frequency is proposed and the performance of this feature has better results considering the standard ones. The whole system performance is evaluated by comparing the input SNR level with the output SNR level. The result shows that the proposed system performance is great, even for the low SNR levels.