Details

Autor(en) / Beteiligte

• Hongdan, Wang
• SalmiJamali, Siti
• Zhengping, Chen
• Qiaojuan, Shan
• Le, Ren

Titel

An intelligent music genre analysis using feature extraction and classification using deep learning techniques

Ist Teil von

• Computers & electrical engineering, 2022-05, Vol.100, p.107978, Article 107978

Ort / Verlag

Amsterdam: Elsevier Ltd

Erscheinungsjahr

2022

Quelle

Access via ScienceDirect (Elsevier)

Beschreibungen/Notizen

• Music genre designations are useful for grouping songs, albums, and performers with comparable musical characteristics into larger categories. The goal of our study and research is to develop a deep learning method that can predict and classify song genres better than existing algorithms. Here the dataset of music genre information has been collected and processed for predicting genre of songs. We present a new approach including feature extraction and classification that takes into account the disparities in spectrums. The dataset namely MSD-I dataset, GTZAN Dataset and ISMIR2004 Genre dataset are utilized for feature extracted using BiLSTM and classification of extracted features has been done using VGG-16 Net. The effect of proposed approach is then evaluated in experiments on single and multi-label genre classification. The results are obtained based on the parameters of accuracy of 97%, precision of 94%, recall of 86.5%, F-1 score of 77.8%, average loss of audio signal of 40% for proposed technique. The audio clip is the only input in the proposed system, and it is processed and features retrieved from it. The retrieved features are then sent into the RNN model's LSTM layer, which produces a trained model for music genre prediction. Initially music dataset is taken and feature vectors are generated using BiLSTM. The generated vectors are passed to the VGG16 NET classification framework that classifies diverse genre of music. The user can then interact with the trained algorithm in order to guess the genre. [Display omitted]