Research on Performance of the Classifying Models Based on Chinese, Pakistani, and Other Genres

1. Introduction

Classification of music or audio seems to be a hot topic nowadays: researchers are trying to acquire knowledge to classify it either audio or music. Music classification plays a vital role in many applications from basic Apps (Application) to complex software. Music information retrieval (MIR) is an interdisciplinary science for fetching data from music. Music information retrieval (MIR) is still in its growing stages but applicable in many real-time appliances. Classification is one of the conspicuous factors of Music information retrieval (MIR). MIR also includes fingerprinting, cover song detection, genre recognition; transcription, recommendation, symbolic melodic similarity, mood, source separation, instrument recognition, pitch tracking, tempo estimation, score alignment, song structure, beat tracking, key detection and query by humming are the main areas of music information retrieval (MIR).

Music classification has gained an attention lately. The genre classification is performed by Panagakis et al. (2009). Li et al. (2003) discussed the automatic music genre classification and new feature extraction has been proposed. Tzanetakis and Cook (2002) have suggested that curtain set of features for modeling a music signals additionally, using features for music classification with the help of K-Nearest Neighbors and Gaussian Mixture models. Lambrou et al. (1998) worked on musical signal classification by using statistical pattern recognition to distinguish three musical genre of rock, piano and jazz. Logan (2000) analyzes Mel frequency cepstral coefficients (MFCCs) for music and speech modeling and use of the discrete cosine transform (DCT) to decorrelate the Mel-spectral vectors. Foote & Uchihashi (2001) introduce a new method of automatic characterization of rhythm and tempo for music and audio. Deshpande et al.(2001) classified music into three categories; rock, classical and jazz by using k-nearest neighbor, Gaussian Mixtures and Support Vector Machines classifiers. Soltau et al. (1998) uses temporal structure modeling and features are abstract features are learned through traditional neural networks for music genre identification. Other methods that are used to classify music through Octave-Based Spectral Contrast and cepstral (MFCC) features by Lee et al.(2009) . Sauders (1996) uses a technique in which speech is separated from music on broadcast FM radio. Bischoff et al. (2009) worked on music classification that can predict the genre and style recommendation. They have compared results of recommended mood with music genre and style. In the recent two to three years numerous amount of work done on genre classification we will discuss a few of them; To recommend the new music for listeners is important in many software applications, approach used by Elbir and Aydin (2020) features are extracted by novel deep neural network. Acoustic features obtained from the network are used for music recommendation. Vishnupriya and Meenakshi (2018) used a deep learning approach to train and classify a system, they use MFCC used as a feature vector for a sound sample and obtain accuracy up to 76% for genre classification. KIKUCHI et al. (2020) utilizes music summarization and compares multiple sections of music with traditional techniques obtain from a single section which gives better results by summarization of music. In this paper Pelchat and Gelowitz (2019) spectrogram images are produced by time-slices of songs, this is used as input for the neural network to classify the songs into genres. Fulzele et al. (2018) work on a hybrid model to increase the accuracy of the GTZAN music dataset. They use Long Short-Term Memory (LSTM) and Support Vector Machine (SVM) models combined.