Genre classification system of TV sound signals based on a spectrogram analysis

In this paper, a genre classification system of TV sound signals is proposed to provide a proper timbre automatically to the listener. The spectral and temporal properties of TV sound are analyzed using the spectrogram principle to classify the genre. Some acoustic features such as the frequency distribution energy variation, etc. are extracted for classification. The sound is first classified into speech or music signals and the music signals are again divided into popular, jazz, or classical mode. After the classification, the frequency gain of the sound is regulated by the existing compensation data according to the genre. The proposed system is implemented using 7 bandpass filters, an 8-bit microprocessor, and a threshold circuit for calculating the silence interval of the signal. From the experimental results, the classification accuracy of speech and music was 95%, and the accuracy between popular, jazz, and classical music was 75%, 30%, and 60%, respectively.