|
Audio information play important role in speaker
identification and semantic based video content
analysis, indexing and retrieval. Sometimes, audio
clues are dominant in determining the story unit types.
In this paper, a new temporal spectral feature
including the proposed spectral histogram is integrated
for audio content classification. By analysis the
temporal spectral distribution, we adaptively determine
the effective feature vectors for audio content
discrimination. Finally, several one class support
vector machine (SVM) are used to classify each audio
clip into following five types: silence, pure music, pure
speech, speech with noise background (Speech+Noise),
and speech with music background (Speech+Music).
Experimental results show the effectiveness of the
proposed methods. |
|
Keywords:Audio classification;SVM;zero-cross rate;bandwidth;spectral histogram |
|