Study of Indian instruments for designing a speech/music classifier
Arvind Kumar, Sandeep Singh Solanki, Mahesh Chandra · 2020 5th International Conference on Computing, Communication and Security (ICCCS) · 2020
Automatic speech/music classification is the art of labeling incoming audio samples into either of two classes i.e. speech and music, using multimedia signal processing technique. This work focuses on building a system model to classify speech and music segments. Speech samples were obtained from standard S&S database and music samples were obtained by playing different Indian musical instruments. Different state-of-the-art features for audio samples proposed in literature were extracted and variation of classification accuracies of these models built on these feature vectors was observed. Various experiments were conducted with Support Vector Machines (SVM) and Random Forest (RF) Classifiers. Best mean classification accuracy of 99.62% and 99.13% is observed for Mel Frequency Cepstral Coefficients (MFCC) feature with SVM and RF classifier respectively.