A new algorithm to Speech/Music discrimination based on gray correlation analysis

Automatic discrimination of speech and music is important in content-based indexing and retrieval of audiovisual media. Previous work concerns with a large number of features like pitch, brightness, variances, time-averages of spectral parameters and the algorithms are not derived from an analysis of the data. In this study, the application of gray correlation analysis algorithm in speech and music discrimination based on unique probability statistics of short energy root mean square (RMS) is present. This algorithm focuses on incomplete feature information, is of what information are part definitude, part ambiguity which has poor information. Experimental results have shown discrimination performance through this approach can be more effective than that of 151 based on one or two features in real-time multimedia applications.