A Hybrid CNN and RNN Variant Model for Music Classification

Applied Sciences, Год журнала: 2023, Номер 13(3), С. 1476 - 1476

Опубликована: Янв. 22, 2023

Music genre classification has a significant role in information retrieval for the organization of growing collections music. It is challenging to classify music with reliable accuracy. Many methods have utilized handcrafted features identify unique patterns but are still unable determine original characteristics. Comparatively, using deep learning models been shown be dynamic and effective. Among many neural networks, combination convolutional network (CNN) variants recurrent (RNN) not significantly considered. Additionally, addressing flaws particular model, this paper proposes hybrid architecture CNN RNN such as long short-term memory (LSTM), Bi-LSTM, gated unit (GRU), Bi-GRU. We also compared performance based on Mel-spectrogram Mel-frequency cepstral coefficient (MFCC) features. Empirically, proposed Bi-GRU achieved best accuracy at 89.30%, whereas hybridization LSTM MFCC 76.40%.

Язык: Английский

---

State-of-the-Art in 1D Convolutional Neural Networks: A Survey

IEEE Access, Год журнала: 2024, Номер 12, С. 144082 - 144105

Опубликована: Янв. 1, 2024

Deep learning architectures have brought about new heights in computer vision, with the most common approach being Convolutional Neural Network (CNN). Through CNN, tasks previously deemed unattainable, including facial recognition, autonomous driving systems, and sophisticated medical diagnostics, among others can now be achieved. layers, non-linear processing units, subsampling layers are used conjunction throughout several phases that make up CNN's structure. Generally, 2D 3D CNNs been to achieve impressive results across numerous areas, survey papers published review their state-of-the-art applications. However, they unsuitable some domain-specific areas where temporal dynamics dependencies must captured. Examples of such domains time series prediction signal identification, which necessitates use one-dimensional signals. Recently, 1D-CNN has evolved develop various models cut research fields. there no paper detailing evolution advancements applications vision tasks. In addressing this gap, provides first exhaustive examine historical development 1D-CNNs elucidate structural intricacies architectural frameworks. It also highlights recent more than twelve distinct domains. Furthermore, an overview significant challenges impacting current training deployment while highlighting potential directions for future exploration. By carrying out survey, researchers fields a comprehensive understanding evolution, intricacies, This equip knowledge needed address faced hurdles.

Язык: Английский

---

Music Deep Learning: Deep Learning Methods for Music Signal Processing—A Review of the State-of-the-Art

IEEE Access, Год журнала: 2023, Номер 11, С. 17031 - 17052

Опубликована: Янв. 1, 2023

The discipline of Deep Learning has been recognized for its strong computational tools, which have extensively used in data and signal processing, with innumerable promising results. Among the many commercial applications Learning, Music Signal Processing received an increasing amount attention over last decade. This work reviews most recent developments processing. Two main that are discussed Information Retrieval, spans a plethora applications, Generation, can fit range musical styles. After review both topics, several emerging directions identified future research.

Язык: Английский

---

DASOD: Detail-aware salient object detection

Image and Vision Computing, Год журнала: 2024, Номер 148, С. 105154 - 105154

Опубликована: Июнь 24, 2024

Язык: Английский

---

Enhanced capsule neural network with advanced triangulation topology aggregation optimizer for music genre classification

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Янв. 2, 2025

Язык: Английский

---

Deep neural networks and fractional grey lag Goose optimization for music genre identification

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Фев. 25, 2025

Generally, music genres have not new established framework, since they are often determined by the composer's background cultural or historical impact and geographical origin. In this work, a methodology is presented based on deep learning metaheuristic algorithms to enhance performance in style categorization. The model consists of two main parts: pre-trained model, ZFNet, through which high level features extracted from audio signals ResNeXt for classification. A fractional-order-based variant Grey Lag Goose Optimization (FGLGO) algorithm used optimize parameters boost model. dual-path recurrent network employed real-time generation evaluate benchmark datasets, ISMIR2004 extended Ballroom, compared state-of-the-art models included CNN, PRCNN, BiLSTM BiRNN. Experimental results show that with accuracy rates 0.918 Ballroom dataset 0.954 dataset, proposed improves efficiency incrementally over existing models.

Язык: Английский

---

Instrument sound classification using a music-based feature extraction model inspired by Mozart's Turkish March pattern

Alexandria Engineering Journal, Год журнала: 2025, Номер 118, С. 354 - 370

Опубликована: Янв. 23, 2025

Язык: Английский

---

Dual attention and channel transformer based generative adversarial network for restoration of the damaged artwork

Engineering Applications of Artificial Intelligence, Год журнала: 2023, Номер 128, С. 107457 - 107457

Опубликована: Ноя. 15, 2023

Язык: Английский

---

Musical Genre Classification using Advanced Audio Analysis and Deep Learning Techniques

IEEE Open Journal of the Computer Society, Год журнала: 2024, Номер 5, С. 457 - 467

Опубликована: Янв. 1, 2024

Classifying music genres has been a significant problem in the decade of seamless streaming platforms and countless content creations. An accurate genre classification is fundamental task with applications recommendation, organization, understanding musical trends. This study presents comprehensive approach to using deep learning advanced audio analysis techniques. In this study, method was used tackle classification. For GTZAN dataset chosen for examines challenge categorization Convolutional Neural Networks (CNN), Feedforward (FNN), Support Vector Machine (SVM), k-nearest Neighbors (kNN), Long Short-term Memory (LSTM) models on dataset. precisely cross-validates model's output following feature extraction from pre-processed data then evaluates its performance. The modified CNN model performs better than conventional NN by capacity capture complex spectrogram patterns. These results highlight how algorithms may improve systems categorizing genres, implications various music-related user interfaces. Up point, 92.7% dataset's correctness achieved 91.6% ISMIR2004 Ballroom

Язык: Английский

---

NSSI questionnaires revisited: A data mining approach to shorten the NSSI questionnaires

PLoS ONE, Год журнала: 2023, Номер 18(4), С. e0284588 - e0284588

Опубликована: Апрель 21, 2023

Non-suicidal self-injury (NSSI) is a psychological disorder that the sufferer consciously damages their body tissues, often too severe requires intensive care medicine. As some individuals hide NSSI behaviors, other people can only identify them if they catch while injuring, or via dedicated questionnaires. However, questionnaires are long and tedious to answer, thus answers might be inconsistent. Hence, in this study for first time, we abstracted larger questionnaire (of 662 items total) own 22 (questions) data mining techniques. Then, trained several machine learning algorithms classify based on into two classes.Data from 277 previously-questioned participants used methods select features highly represent NSSI, then 245 different were asked participate an online test validate those methods.The highest accuracy F1 score of selected features-via Genetics algorithm-are 80.0% 74.8% respectively Random Forest algorithm. Cronbach's alpha (validation features) 0.82. Moreover, results suggest MLP classes Positive Negative with 83.6% 83.7% F1-score questions.While previously psychologists many combined see whether someone involved various methods, present showed questions enough predict not. Then utilized among which, 10 hidden layers had best performance.

Язык: Английский

---