The Application Research Of PCA In Speech Detection

It is essential for speech processing system to have robust speech detection. A PCA(principal component analysis)-based speech detection was proposed. We get a very good result form the examination of using this method.We analyses the different result of PCA by using autocorrelation matrix and covariance matrix, and point out that the express of PCA is different but the error are the same. By research we know that the reason of the difference is the expectation of the samples. From the point view of high dimensional data processing, we get the essence of PCA, and also easy to explain this difference between autocorrelation matrix method and covariance matrix method.In processing asunder distributed data, a fast PCA algorithm was proposed. Without large matrix calculate and no iteration, the fast PCA algorithm gets a very fast speed. It should be pointed out that the result form this algorithm is not the best one of PCA. But the error caused by this algorithm is very small and acceptable.The result of fast PCA is the bass of the new subspace. By analysis the distribute of the data in subspace, the speech and non-speech can be detected. Creating many subspaces to get a batter performance of the detection can handle the variety of the non-speech.