Study of speech/music classification for Kannada language
Arvind Kumar, Faiza Tasneem, Ashutosh Anand, S Natya, Akshaya M Ganorkar, Rakesh Chowdhury, Mahesh Chandra · AIP conference proceedings · 2024
Speech/music classification (SMC) is mostly a two-stage process where the initial stage segments the audio signal based on their statistical characteristics, and the second stage labels these segments as speech or music based on their feature characteristics.We aim to develop a speech/music classifier for the Kannada language and observe the efficiency of the existing state-of-art features on the new Indian speech corpus, and compare the results with existing literature.Since most of the work in this domain is focused on Western languages, this work will open new research doors in different regional languages of India.Experiments were conducted on eight different features using a Support Vector Machine (classifier).Best-performing features were appended to form hybrid features.The best classification accuracy of 89.37% is seen for the combination of short-term energy and spectral flux features.FIGURE 1. Speech waveform and its normalized form