Abstract :
The Automatic Annotation of a Singer in a Video Song System enables the user to search for their favorite
Singer’s video song from video store. The proposed methodology performs the search in a video store by comparing
the content of the video rather than the user’s textual query and tags associated with the videos. We make an effort
for identifying underlying Singer in the video songs by mining their audio features, frequencies and onset values.
Various algorithms, filters and classifiers are used to implement this system, namely to mention a few Chebychev
infinite-impulse response(IIR) filter , inverse comb filter, Naive Bayes, Sequential Minimal Optimization (SMO)
Classification Algorithms etc. This paper focused on the extraction of information about the video contents
automatically. The extracted information can serve as an initial step for various data access methods such as surfing,
searching, comparison, and classification. It is worth mentioning that annotating music information in a video is an
emerging task and was not much covered in past research papers. In the proposed System three Singers namely
S.P. Balasubramaniam, Susheela and Swarnalatha are selected for analysis. For each Singer 100 video songs of
length 10 seconds duration are taken. From these video songs the vocal track alone are extracted by using IIR digital
filter and inverse comb filter. Mathematical functions are applied to calculate the Spectral and Cepstral features from
the extracted signal. These features are applied to five classifiers for classification. This System gives a maximum
of 95% accuracy in identifying a Singer in a video song using SMO classifier.
Keyword :
Content Based Search, Video annotation, Singer Identification, Spectral Features, Mel Frequency Cepstral Coefficients.