This paper introduces a new class of nonlinear adaptive filters, whose structure is based on Hammerstein model. Such filters derive from the functional link filter (FLAF) model, defined by input expansion, which enhances representation signal through projection in higher dimensional space, and subsequent filtering. In particular, two robust FLAF-based architectures are proposed designed ad hoc to tackle nonlinearities acoustic echo cancellation (AEC). The simplest architecture split FLAF,...

The extreme learning machine (ELM) was recently proposed as a unifying framework for different families of algorithms. classical ELM model consists linear combination fixed number nonlinear expansions the input vector. Learning in is hence equivalent to finding optimal weights that minimize error on dataset. update works batch mode, either with explicit feature mappings or implicit defined by kernels. Although an online version has been former, no work done up this point latter, and whether...

Breast cancer is the most widespread neoplasm among women and early detection of this disease critical. Deep learning techniques have become great interest to improve diagnostic performance. However, distinguishing between malignant benign masses in whole mammograms poses a challenge, as they appear nearly identical an untrained eye, region (ROI) constitutes only small fraction entire image. In paper, we propose framework, parameterized hypercomplex attention maps (PHAM), overcome these...

In this paper two novel nonlinear cascade adaptive architectures, here called sandwich models, suitable for the identification of general systems are presented. The proposed architectures rely on combination structural blocks, each one implementing a linear filter or memoryless function. All functions involved in adaptation process based spline and can be easily modified during learning using gradient-based techniques. particular, simple form on-line algorithms is derived. addition, we...

The L3DAS22 Challenge is aimed at encouraging the development of machine learning strategies for 3D speech enhancement and sound localization detection in office-like environments. This challenge improves extends tasks L3DAS21 edition. We generated a new dataset, which maintains same general characteristics datasets, but with an extended number data points adding constrains that improve baseline model's efficiency overcome major difficulties encountered by participants previous challenge....

The modeling of human emotion expression in speech signals is an important, yet challenging task. high resource demand recognition models, combined with the general scarcity emotion-labelled data are obstacles to development and application effective solutions this field. In paper, we present approach jointly circumvent these difficulties. Our method, named RH-emo, a novel semi-supervised architecture aimed at extracting quaternion embeddings from real-valued monoaural spectrograms, enabling...

Semantic communications represent a significant breakthrough with respect to the current communication paradigm, as they focus on recovering meaning behind transmitted sequence of symbols, rather than symbols themselves. In semantic communications, scope destination is not recover list identical ones, but message that semantically equivalent emitted by source. This paradigm shift introduces many degrees freedom encoding and decoding rules can be exploited make systems much more efficient....

Recently, a new class of nonlinear adaptive filtering architectures has been introduced based on the functional link filter (FLAF) model. Here we focus specifically split FLAF (SFLAF) architecture, which separates adaptation linear and coefficients using two different filters in parallel. This property makes SFLAF well-suited method for problems like acoustic echo cancellation (NAEC), separation tasks brings some performance improvement. Although flexibility is one main features SFLAF,...

Recently, a novel class of nonlinear adaptive filters, called spline filters (SAFs), has been introduced and demonstrated to be very effective in many practical applications. The learning rules these architectures are based on the least mean square (LMS) algorithm. In order provide theoretical foundation SAF, this paper we steady-state performance evaluation. particular, after stochastic analysis behavior SAF approach under Gaussian assumption, analytical derivation excess error (EMSE)...

Learning from data in the quaternion domain enables us to exploit internal dependencies of 4D signals and treating them as a single entity. One models that perfectly suits with quaternion-valued processing is represented by 3D acoustic their spherical harmonics decomposition. In this paper, we address problem localizing detecting sound events spatial field using processing. particular, consider harmonic components captured first-order ambisonic microphone process convolutional neural...

Hypercomplex neural networks have proven to reduce the overall number of parameters while ensuring valuable performance by leveraging properties Clifford algebras. Recently, hypercomplex linear layers been further improved involving efficient parameterized Kronecker products. In this article, we define parameterization convolutional and introduce family (PHNNs) that are lightweight large-scale models. Our method grasps convolution rules filter organization directly from data without...

Semantic communication is expected to be one of the cores next-generation AI-based communications. One possibilities offered by semantic capability regenerate, at destination side, images or videos semantically equivalent transmitted ones, without necessarily recovering sequence bits. The current solutions still lack ability build complex scenes from received partial information. Clearly, there an unmet need balance effectiveness generation methods and complexity information, possibly taking...

In this paper, we propose a Deep Recurrent Neural Network (DRNN) approach based on Long-Short Term Memory (LSTM) units for the classification of audio signals recorded in construction sites. Five classes multiple vehicles and tools, normally used sites, have been considered. The input provided to DRNN consists concatenation several spectral features, like MFCCs, mel-scaled spectrogram, chroma contrast. proposed architecture feature extraction described. Some experimental results, obtained by...

The L3DAS21 Challenge is aimed at encouraging and fostering collaborative research on machine learning for 3D audio signal processing, with particular focus speech enhancement (SE) sound localization detection (SELD). Alongside the challenge, we release dataset, a 65 hours corpus, accompanied Python API that facilitates data usage results submission stage. Usually, approaches to tasks are based single-perspective Ambisonics recordings or arrays of single-capsule microphones. We propose,...

While deep generative models are showing exciting abilities in computer vision and natural language processing, their adoption communication frameworks is still far underestimated. These methods demonstrated to evolve solutions classic problems such as denoising, restoration, or compression. Nevertheless, can unveil real potential semantic frameworks, which the receiver not asked recover sequence of bits used encode transmitted (semantic) message, but only regenerate content that...

Directly sending audio signals from a transmitter to receiver across noisy channel may absorb consistent bandwidth and be prone errors when trying recover the transmitted bits. On contrary, recent semantic communication approach proposes send semantics then regenerate semantically content at without exactly recovering bitstream. In this paper, we propose generative framework that faces problem as an inverse problem, therefore being robust different corruptions. Our method transmits...

Deep probabilistic generative models have achieved incredible success in many fields of application. Among such models, variational autoencoders (VAEs) proved their ability modeling a process by learning latent representation the input. In this paper, we propose novel VAE defined quaternion domain, which exploits properties algebra to improve performance while significantly reducing number parameters required network. The proposed with respect traditional VAEs relies on leverage internal...

This paper addresses the problem of translating night-time thermal infrared images, which are most adopted image modalities to analyze scenes, daytime color images (NTIT2DC), provide better perceptions objects. We introduce a novel model that focuses on enhancing quality target generation without merely colorizing it. The proposed structural aware (StawGAN) enables translation better-shaped and high-definition objects in domain. test our aerial DroneVeichle dataset containing RGB-IR paired...