Content-Based Audio Classification and Retrieval for Audiovisual Data Parsing

<title>Semi-automatic approach for music classification</title>

“…without human voice). Zhang and Kuo proposed a method in [1] to classify an audio signal into a number of basic audio types, including pure instrumental music (denoted as "pure music" in [1]) and singing (denoted as "song" in [1]). While instrumental music is normally indicated by stable frequency peaks in the spectrogram, i.e.…”

“…Compared to the ZCR of instrumental music which mostly stays within a relatively small range of amplitude, there are high peaks in the ZCR of vocal music due to the pronunciation of some consonant components. In [1], pure instrumental music was distinguished from other types of music, including singing, based on the average zero-crossing rate and the fundamental frequency properties. Four aspects were checked: the degree of being harmonic, the degree of the fundamental frequency's concentration on certain values during a period of time, the variance of the average zero-crossing rates, and the range of amplitudes of the average zero-crossing rates.…”

“…Two kinds of audio features are checked. On one hand, the frequency level of spectral peak tracks is checked in the frequency domain (for definition and calculation of spectral peak tracks, please refer to [1]). If frequencies of a large percentage of the spectral peak tracks change significantly over time, i.e.…”

“…In contrast, for tracks in a chorus, the derivative values are mostly around zero. On the other hand, a method for detecting significant peaks in a spectrum is proposed in [1]. This method can be applied to check whether there are significant harmonic partials in higher frequency bands of the signal, and decide whether it is a solo song or chorus.…”

“…This can be done by checking the range of pitch values in the song. There are various ways of estimating the pitch or the fundamental frequency (which is equivalent to pitch) of speech and music signals in the literature [1], [4] . The pitch of the singer's voice is estimated once every certain period of time along the song, e.g.…”