Building a computational model for mood classification of music by integrating an asymptotic approach with the machine l
- PDF / 1,241,496 Bytes
- 13 Pages / 595.276 x 790.866 pts Page_size
- 9 Downloads / 157 Views
ORIGINAL RESEARCH
Building a computational model for mood classification of music by integrating an asymptotic approach with the machine learning techniques Sanchali Das1,3 · Bidyut K. Bhattacharyya1,2 · Swapan Debbarma1,3 Received: 4 March 2020 / Accepted: 22 May 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract In this paper, we are working to understand the statistical behavior of acoustic features of audio, for one of the low resource languages, which is Kokborok from North East India. First, we have developed a classification system for Kokborok music by using the traditional machine learning technique. We used mainly Timbre, Rhythm, and Intensity feature to classify songs between four classes having three subclasses. This classification system gives poor performance compared to other Indian languages and western languages. So, we develop a computational method to minimize the errors for each class for the overall system. For such poor low resource language, the ground truth set creation is very tough. So, the behavior of the audio features of each song is analyses mathematically to understand whether the truth set is correct or not. Technically the feature values have to be different for each class and in a similar range for subclasses. We have defined a statistical parameter called “alpha” (α), for estimating the better value of the accuracy rate. This parameter alpha eventually estimates the final accuracy rate. This alpha is calculated, and the final value of the accuracy rate was calculated by extrapolating when the number of songs goes to infinity. The method enhances the actual accuracy rate from 49 to 63%, in the limit when the number of samples goes to infinity. Overall, our approach, when used in conjunction with the machine learning method, can predict a better accuracy rate for Kokborok music. Keywords Computational data analysis · Mood classification · Statistical modeling of music mood · Theoretical analysis of music mood · Kokborok music
1 Introduction Music information retrieval (MIR) community is closely related to information retrieval (IR), and in recent years, it is in its evaluation phase. For IR, ground truth creating is costly for query related applications so that the efficiency of ground truth annotation is directly proportional to evaluation (Urbano et al. 2013). The annotation procedure might not always get an accurate result because the truth data set had its component. Due to the distribution of the number of * Sanchali Das [email protected] 1
National Institute of Engineering and Technology, Agartala, India
2
Electrical Engineering Department, Agartala, India
3
Department of Computer Science and Engineering, NIT Agartala, Jirania, Agartala, India
songs in different categories, manual annotation generates errors for a given class. So, it is essential to have sample sizes in each group nearly the same or in equal size, which is very difficult. Thus, we need to have proper test data collection, which leads to almost even distribut
Data Loading...
