No abstract

In this paper, we propose new dereverberation method for underwater speech communication. The speech signal generated by a fixed one source. Then the signal received on 9 horizontal sensor array in shallow water environment. Finally, received signals are processed dereverberation. The received signals achieved delay-and-sum beamforming and then, signals calculates time delay both each signals. Next, we calculate Hilbert envelope using time delay information and LP residuals. At last, we acquire new LP residuals using various parameters. Proposed technique shows better performance than previous dereverberation technique. I. INTRODUCTIONFor the past decade, there has been a tremendous increase in research and development of underwater acoustic communication systems. However, many problems exist there. The performance of coherent acoustic communication systems is highly dependent upon channel characteristics including multipath, spatial and temporal coherence, Doppler effects and ambient noise. An important feature of the process of the sound field formation in oceanic waveguides is the interference caused by the surface and bottom reflections and by the layered structure of the water bulk. This means multipath effect. The interference effects lead to the reverberation, which consists in the fact that, in addition to the main signal, multiple signals are received that propagate over different paths and arrive at the receiver with different delays. The effect of reverberation on speech is to cause it to sound distant and spectrally distorted and can also reduce intelligibility. Dereverberation is therefore an important speech enhancement process for underwater communication.Several dereverberation algorithms proposed by various authors. They have been proposed and can be considered in two categories: (i) algorithms based on processing of the linear prediction (LP) residual and (ii) blind channel estimation/inversion algorithm. In this paper, we adopted method based on LP residual [ 1,2]. Several LP residual processing techniques have been developed using established models of speech production. Some method use wavelet extrema clustering to reconstruct an enhanced LP residual. B. Yegnanarayana et al derive a weighting function based on the direct-to-reverberant ratio in different regions of the LP