Feature Extraction for an Audio Discrimination between Speech and Music for Better Human and Computer Interaction
7 Pages Posted: 21 Jan 2021
Date Written: January 20, 2021
In today's world, where everyone wants everything just by saying, therefore audio plays a important role. The focus of the paper is demand driven with a need to discriminate between music and speech. The paper also focuses on proposing a related feature for retrieval of information, human interface with machine and editing of the video. Here is an attempt where a model is simulated using average cepstrum, wherein the human consciousness has the tendency of picking the cepstral changes that are larger. The data that is cepstrum will be reduced in magnitude exponentially if it is away from the mean value. In this paper, experiments are performed for comparing the classification of the feature between the proposed feature and that with the previously proposed feature. This type of dynamic classification demonstrates that the test database for proposed features is having good quality of music and speech classification.
Keywords: Restricted Boltzmann Algorithm, Monte Carlo Algorithm, Cepstral Coefficient, Linear Predictive Coding, Spectral Flux, Spectral Centroid, Contrastive Divergence Learning
Suggested Citation: Suggested Citation