Stacked auto-encoders based visual features for speech/music classification

Deep Learning has emerged as a powerful set of computational methods achieving great results in a variety of different tasks. Music signal processing, a field with rich commercial applications, seems to benefit too from this data-driven approach. In this paper a review of the state of the art Deep Learning methods applied on music signal processing is provided. A special focus is given in music information retrieval and music generation. In addition, possible future research directions are discussed.

This paper describes the construction of a system called i-Ring that can generate a polyphonic ringtone based on a user's humming input. Algorithms used in the system for music transcription and chord generation include various techniques in speech and music signal processing, such as pitch tracking and dynamic programming, which are explained in the paper. Experimental results demonstrate the feasibility of i-Ring as a handy tool for computer assisted music composition for polyphonic ringtones

This thesis presents a set of works on the use of deep learning (DL) in the next generation of communications systems. Digital communications is a cornerstone technology of the modern information age. Thus, a natural question to ask is whether DL enables better solutions for common problems in communications systems. To that end, this thesis poses two research questions: RQ1: How can deep learning techniques be effectively applied across different parts of the networking stack? RQ2: What are the advantages and challenges of implementing deep learning in the physical layer versus the application layer? This thesis consists of two parts with three chapters each. The first three content chapters discuss works that relate to the physical layer of the networking stack. The second three chapters contain works that relate to the application layer of the networking stack. Chapter 2 focuses on improving deep learning in the physical layer of communication networks. We develop a more flexible neural receiver that can handle multiple types of modulations. A modulation type is a configuration that makes a trade-off between speed and reliability. This approach makes deep learning based communication systems modular, making them more adaptable and efficient. Chapter 3 continues work on the physical layer, tackling the challenge of making deep learning receivers faster and more energy-efficient. We use a method of deep learning called group equivariant deep learning to build neural networks that inherently understand certain properties of radio signals. This results in smaller, more efficient networks that perform just as well as larger ones. Chapter 4 shifts to underwater acoustics. We use a technique called contrastive learning to train neural networks without labeled data. This approach could improve underwater sound classification and potentially help simulate underwater communication channels more accurately. Chapter 5 explores a security vulnerability in video streaming platforms like YouTube. Videos are generally streamed dynamically which is known to result in vulnerabilities. Using deep learning techniques, specifically deep-metric learning, we demonstrate that video IDs can be identified from encrypted streams with minimal examples. This work highlights how DL simplifies exploiting vulnerabilities between communication layers. Chapter 6 focuses on improving neural image compression. Building upon the work of Yang et al. (2020), we propose SGA+, a set of functions that improve the weighting of Gumbel probabilities. Our approach, particularly the SSL function with hyperparameter a, converges faster and achieves better PSNR/BPP trade-offs than the original SGA method. This advancement in neural image compression is important for reducing internet congestion and improving edge device capabilities. Chapter 7 investigates the application of Federated Learning (FL) in Face Recognition (FR) to address privacy concerns and issues with data heterogeneity. Given the challenges of non-shared identities across parties, we propose using federated meta-learning. Our approach demonstrates improved overall performance per client, especially under heterogeneous data splits. Notably, the performance gains primarily benefit weaker clients, reducing the variance in performance across clients. Together, these works provide partial answers to the research questions posed above. Major factors in applying DL across network layers are modularity, translating DL research to communications, and integrating DL advancements into the communications stack of tomorrow. Notably, there are differences between the application layer and physical layer. This thesis found the physical layer to benefit more broadly from the inclusion of DL. However, the models used throughout this work were highly similar; they were all based on convolutional neural networks (CNNs). The future of DL for communications looks bright and much work remains to be done.

Music signal processing is one of the active research area now a days. Identifying the musical instrument form the solo recordings is one of the applications under signal processing. Different types of algorithms have been proposed till now to identify the instrument. For this objective first of all we have to extract the features from sound samples and then we can use them for instrument identification. In this paper we are presenting one of efficient method to extract the features from sound sample that is Mel Frequency Cepsrtal Coefficient (MFCC). These MFCCs are closely related to behavior of human auditory system. While calculating these MFCCs we actually deal with spectral envelop. And hence it gives very distinctive features of the sound samples. So using these features for identification of musical instrument we can increase the accuracy of identification. In our work we have used five instruments and 30 samples of each instrument. In the training database total 90 samples (60% of total samples) and for testing database 60 (40%)samples are used. In the classification phase we have used K-Nearest Neighbor (K-NN) classifier. Proposed system gives the accuracy of 91.66% for Cello, Piano and Trumpet and accuracy of 83.33% for Flute and Violin.

ABSTRACTIn recent years, there has been a significant surge in the utilisation of deep learning and machine learning techniques for addressing complex and time‐intensive problems. The significance of employing deep learning becomes increasingly evident as the complexity of the problem increases. In the field of electromagnetics, the utilisation of deep learning techniques has exhibited exceptional efficacy across many applications, especially in wireless communications. In wireless communications, providing a structure that can simultaneously generate multiple beams and beamform them involves complexity and specific constraints. In this article, the time modulation technique is utilised to generate harmonics in the sidebands alongside the fundamental beam in various planar antenna arrays. By demonstrating the nonlinear shaping of the harmonic beams relative to each other, a novel approach is proposed that leverages deep learning techniques for the beamforming of both the fundamental beam and harmonic beams. In this regard, two models are proposed: a deep neural network (DNN) and a convolutional neural network (CNN). The input of CNN is comprised of two‐dimensional patterns of the main beam and harmonics. The input to DNN, on the other hand, includes useful details about the main beam and harmonics, such as their scanning angles, side lobe levels and directivities. The output of the models consists of the time modulation parameters of the array elements, including the pulse width and the pulse delay. The results demonstrate that DNN has achieved better accuracy and a shorter processing time in comprehending the relationship between the time modulation of array elements with different array dimensions and the radiation pattern of the fundamental beam and harmonic beams. Additionally, several samples are presented to evaluate the proposed model. The results demonstrate a high level of accuracy in fundamental beamforming, as well as in harmonic beamforming and beam steering.

Designing of VehiNet Using Convolutional Neural Networks and Deep Learning Techniques

This chapter takes a new look at an old riddle by discussing putative parallelisms between speech and music processing, as well as music to speech transfer effects. Interestingly, both speech and music rely on comparable acoustic information (i.e. timing, pitch, and timbre) and share, at least partially, perceptive, cognitive, motivational, and motor components. Importantly, in this context, the chapter introduces a cortical framework of speech processing, as well as the perceptual and cognitive operations underlying speech learning mechanisms. It then reviews previous literature pinpointing the influence of music training on segmental and suprasegmental aspects of speech processing as well as on cognitive functions. Two complementary hypotheses underlying such transfer effects—cascade effects and multidimensional processes—are proposed. The issue of shared versus distinct neural networks for music and speech processing is also discussed. Finally, the chapter integrates the often observed perceptual and cognitive advantages of musicians in a holistic framework by considering their interrelations during an ecological, valid, word-learning task.

In today’s digital world, Android phones play a vital part in a variety of facets of both professionals and individuals’ personal and professional lives. Android phones are great for getting things done faster and more organized. The proportionate increase in the number of malicious applications has also been seen to be expanding. Since the play store offers millions of apps, detection of malware apps is challenging task. In this paper, a methodology is introduced for detecting malware in Android applications through the utilization of global image shape transform (GIST) features extracted from grayscale images of the applications. The dataset comprises samples of both malware and benign apps collected from the virus share website. After converting the apps into grayscale images, GIST features are extracted to capture their global spatial layout. Various machine learning (ML) algorithms, such as logistic regression (LR), k-nearest neighbour (KNN), AdaBoost, decision tree (DT), Naïve Bayes (NB), random forest (RF), support vector machine (SVM), extra tree classifier (ETC), and gradient boosting (GB), are employed to classify the applications according to their GIST features. Furthermore, a feed forward neural network (FFNN) is utilized as a deep learning (DL) technique to further improve the accuracy of classification. The performance of each algorithm is evaluated using metrics such as accuracy, precision and recall. The results demonstrated that the FFNN achieves superior accuracy compared to traditional ML classifiers, indicating its effectiveness in detecting malware in Android apps.

Purpose: Music is an artistic expression with many categories in various genres and styles, characterized by its melodic and harmonic compositions. Music genre classification is crucial because genres serve as descriptors commonly used to organize large music collections, especially on the internet and in widely used applications like JOOX and Spotify. The aim of this research is to implement the Mel Frequency Cepstral Coefficients (MFCC) feature extraction method to generate numerical features from a set of specific music tracks. This collection of information will then be classified using machine learning. Methods: The method used in this study begins with combining the "GTZAN Dataset - Music Genre Classification" with additional data from TikTok and YouTube. The total dataset consists of 1,200 audio files, divided into 12 classes. The MFCC extraction process generates numerical representations of acoustic characteristics, which are then processed using Artificial Neural Networks. Result: The experiments demonstrate that increasing the amount of data is crucial, as it can enhance both variation and accuracy. The average accuracy achieved in this study is 91.42%, while the highest accuracy reaches 92.16%. These findings indicate that this study outperforms previous studies. Novelty: The novelty of this research lies in the integration of dynamic social media data (TikTok and YouTube) to enrich the standard GTZAN dataset, the repetition of the MFCC feature extraction process, and the combination of MFCC with Artificial Neural Networks (ANN).

Music genre classification has become a big task in the age of seamless music streaming services and content development. Accurate music genre classification is critical for applications such as music recommendations, content organization and identifying musical trends. This study proposes a comprehensive approach to music genre classification based on deep learning and advanced audio analysis tools. To address the challenge of accurately classifying music genres for music apps, a deep learning model is used with the GTZAN dataset for classification. The study investigates on various algorithms like Convolutional Neural Networks (CNN), Feedforward Neural Networks (FNN), Support Vector Machines (SVM), k-Nearest Neighbor (kNN) and Long Short-term Memory (LSTM) on this dataset. The study involves cross-validating the models output after feature extraction from the preprocessed audio data and evaluating its performance. This paper proposes an enhanced CNN model which outperforms regular neural networks by taking advantage of its capacity to record complex spectral patterns. These findings show how deep learning algorithms can improve systems for categorizing music genres, potentially benefiting music-related apps and interfaces. Hence, the proposed model has given an accuracy of 94.2% on the GTZAN dataset and 93.3% on the ISMIR2004 Ballroom dataset.

To classify the texts accurately, many machine learning techniques have been utilized in the field of Natural Language Processing (NLP). For many pattern classification applications, great success has been obtained when implemented with deep learning models rather than using ordinary machine learning techniques. Understanding the complex models and their respective relationships within the data determines the success of such deep learning techniques. But analyzing the suitable deep learning methods, techniques, and architectures for text classification is a huge challenge for researchers. In this work, a Contiguous Convolutional Neural Network (CCNN) based on Differential Evolution (DE) is initially proposed and named as Evolutionary Contiguous Convolutional Neural Network (ECCNN) where the data instances of the input point are considered along with the contiguous data points in the dataset so that a deeper understanding is provided for the classification of the respective input, thereby boosting the performance of the deep learning model. Secondly, a swarm-based Deep Neural Network (DNN) utilizing Particle Swarm Optimization (PSO) with DNN is proposed for the classification of text, and it is named Swarm DNN. This model is validated on two datasets and the best results are obtained when implemented with the Swarm DNN model as it produced a high classification accuracy of 97.32% when tested on the BBC newsgroup text dataset and 87.99% when tested on 20 newsgroup text datasets. Similarly, when implemented with the ECCNN model, it produced a high classification accuracy of 97.11% when tested on the BBC newsgroup text dataset and 88.76% when tested on 20 newsgroup text datasets.

The medical diagnosis of diseases involves the detection and classification of patients from medical image data. It has an important Artificial Intelligence (AI) application in Healthcare and Neuroscience fields. In addition, Machine Learning (ML) and Deep Learning (DL) techniques provide important tasks on Neurodegenerative Diseases (ND) diagnosis from Brain MRI (Magnetic Resonance Imaging) data for image processing with good classification results. This chapter aims at comprehensive and comparative study on the performance of medical image classification with those techniques using Brain MRI datasets. We introduce a brief overview of the medical diagnosis of ND with AI applications in the neuroscience domain. In the literature part, we present the understanding of the medical image classification with Machine and Deep Learning techniques. Therefore, we describe the concept of MRI data and the Brain MRI datasets that are used in this work. In the comparison part, we focus on related works of the Artificial Neural Network (ANN) and the Deep Convolutional Neural Network (DCNN) models and we compare these models in the literature. In the methodology part, we combine the ANN and DCNN in order to propose a new Deep model named ANN-DCNN model when it is more accurate and effective to classify from those Brain images. To improve the performance of those Neural Networks in medical diagnosis applications, we discuss the recent insights and future directions.

Understanding human action from the visual data is an important computer vision application for video surveillance, sports player performance analysis, and many IoT applications. The traditional approaches for action recognition used hand-crafted visual and temporal features for classifying specific actions. In this paper, we followed the standard deep learning framework for action recognition but introduced channel and spatial attention module sequentially in the network. In a nutshell, our network consists of four main components. First, the input frames are given to a pre-trained CNN for extracting the visual features and the visual features are passed through the attention module. The transformed features maps are given to the bi-directional LSTM network that exploits the temporal dependency among the frames for the underlying action in the scene. The output of bi-direction LSTM is given to a fully connected layer with a softmax classifier that assigns the probabilities to the actions of the subject in the scene. In addition to cross-entropy loss, the marginal loss function is used that penalizes the network for the inter action classes and complimenting the network for the intra action variations. The network is trained and validated on a tennis dataset and in total six tennis players' actions are focused. The network is evaluated on standard performance metrics (precision, recall) promising results are achieved.

The traditional computer-aided design (CAD) software is often not realistic in rendering scenes, such as light, materials, shadows, and other details, which leads designers to spend a lot of energy on manual adjustment.The rendering algorithm based on deep learning (DL) can automatically learn the complex output, and the effect is usually better than traditional methods.In this article, a convolutional neural network (GA-CNN) based on genetic algorithm (GA) optimization is proposed, and it is combined with CAD data and applied to the rendering of environmental art design scenes.This method uses GA to optimize the parameters of CNN to improve its learning effect and generalization ability.From the overall trend, the response speed of GA-CNN algorithm and support vector machine (SVM).The response speed of the traditional CNN method is relatively slow, and the fluctuation range is large.This is because traditional CNN needs a lot of convolution operations and feature extraction when dealing with artistic scenes in complex environments, which leads to a long calculation time.In practical application, it is necessary to select appropriate algorithms and parameter settings according to specific task requirements to achieve the best results.

Handcrafted features and late fusion with deep learning for bird sound classification