\r\nhotspot in the field of signal processing in recent years. It has

\r\nmany applications and influences in teleconferencing, hearing aids,

\r\nspeech recognition of machines and so on. The sounds received are

\r\nusually noisy. The issue of identifying the sounds of interest and

\r\nobtaining clear sounds in such an environment becomes a problem

\r\nworth exploring, that is, the problem of blind source separation.

\r\nThis paper focuses on the under-determined blind source separation

\r\n(UBSS). Sparse component analysis is generally used for the problem

\r\nof under-determined blind source separation. The method is mainly

\r\ndivided into two parts. Firstly, the clustering algorithm is used to

\r\nestimate the mixing matrix according to the observed signals. Then

\r\nthe signal is separated based on the known mixing matrix. In this

\r\npaper, the problem of mixing matrix estimation is studied. This paper

\r\nproposes an improved algorithm to estimate the mixing matrix for

\r\nspeech signals in the UBSS model. The traditional potential algorithm

\r\nis not accurate for the mixing matrix estimation, especially for low

\r\nsignal-to noise ratio (SNR).In response to this problem, this paper

\r\nconsiders the idea of an improved potential function method to

\r\nestimate the mixing matrix. The algorithm not only avoids the inuence

\r\nof insufficient prior information in traditional clustering algorithm,

\r\nbut also improves the estimation accuracy of mixing matrix. This

\r\npaper takes the mixing of four speech signals into two channels as

\r\nan example. The results of simulations show that the approach in this

\r\npaper not only improves the accuracy of estimation, but also applies

\r\nto any mixing matrix.","references":null,"publisher":"World Academy of Science, Engineering and Technology","index":"Open Science Index 150, 2019"}