Audio MFCC-gram Transformers for respiratory insufficiency detection in COVID-19

This work explores speech as a biomarker and investigates the detection of respiratory insufficiency (RI) by analyzing speech samples. Previous work [Casanova et al. 2021] constructed a dataset of respiratory insufficiency COVID-19 patient utterances and analyzed it by means of a convolutional neural network achieving an accuracy of 87.04%, validating the hypothesis that one can detect RI through speech. Here, we study how Transformer neural network architectures can improve the performance on RI detection. This approach enables construction of an acoustic model. By choosing the correct pretraining technique, we generate a self-supervised acoustic model, leading to improved performance (96.53%) of Transformers for RI detection.

Optimization of Noise Transfer Path Based on the Composite Panel Acoustic and Modal Contribution Analysis

The noise of a cab directly affects the comfort and labor efficiency of the operators. The optimization of the structure-borne transmission path can obviously reduce the cab noise. The method of panel acoustic contribution analysis (PACA) is used to reduce structure noise. However, most studies only consider the panel acoustic contribution of a single frequency, without considering the contribution of major frequencies synthesis to confirm the optimized panels. In this paper, a novel method is proposed based on composite panel acoustic and modal contribution analysis and noise transfer path optimization in a vibro-acoustic model. First, the finite element model (FEM) and the acoustic model are established. Based on the acoustic transfer vector (ATV) method, a composite panel acoustic contribution analysis method is proposed to identify the panels affecting the noise of the field point. Combined with the modal acoustic contribution of the modal acoustic transfer vector (MATV) method, the noise field point is confirmed in the area which has the most significant influence. Second, the optimization algorithm NLOPT which is a nonlinear optimization is applied to design the areas. The noise transfer path optimization with vibroacoustic coupling response can quickly determine the optimal thickness of the panels and reduce low-frequency noise. The effectiveness of the proposed method is applied and verified in an excavator cab. The sound pressure level (SPL) the driver’s right ear (DRE) decreased obviously. The acoustic analysis of the composite panel acoustic contribution and modal acoustic contribution can more accurately recognize an optimized area than the traditional PACA. This method can be applied in the optimization of the structure-borne transmission path for construction machinery cab and vehicle body.

3D CFD Simulation of Phase Resonance in a Pump Turbine with an Acoustic Model

In order to understand the complex nature of the system dynamic phenomena, such as the strong vibration and noise caused by blade passage in the pump turbine, a state-of-the-art three-dimensional (3D) compressible transient simulation would be desirable to study the problem in depth. This study investigated the phase resonance (PR) that occurred during a full-load operation in the turbine mode of a pump turbine on a prototype scale. As a first step, the wave reflection at the boundaries, and the influence of the timestep and sound speeds on the behavior of traveling pressure waves inside a spiral casing, were studied. It was found that nonreflective boundary conditions and an appropriately small timestep are critical to capturing the wave reflection and superposition process inside a spiral casing; a certain kind of direct PR risk was detected in its system design. The detected direct PR differed from the well-known PR with two features: firstly, it was almost independent of the sound speeds, and secondly, the pressure distribution over the spiral circumference varied among the amplitudes. The latter feature was caused by pressure waves at every stator channel induced by a rotor stator interaction (RSI). The 3D flow simulation with an acoustic model, which couples the RSI and PR phenomena, would predict better results for understanding the problem than the simplified one-dimensional (1D) method.

OVERTON-RESONANT SINGING BASIS OF COMPOSITOR LANGUAGE (OVERTONE-RESONANT BASIS OF COMPOSITOR LANGUAGE)

The article substantiates the embodiment of the traditional overtone intonational song beginning in the composer’s work. The application of the main acoustic features of an authentic timbre is described, such as compression, the presence of «specific resonances», reliance on the vocal position and inharmonic overtones. The main composer’s task is to create a piece of music that would imitate or reproduce the resonance characteristic of the folk singing tradition — the overtone. The research is based on the method of sound-acoustic measurement of an audio signal. The work was performed using the Reaper V4,581 audio editor software (Digital Audio Workstation) and a spectrograph built into the Reaper brand iZotop Ozon5 Advansed. This model of the spectrograph allows you to see the spectrogram of the investigated audio material in 3D format and color image, which more accurately reflects the measurements of the vibration frequency of the sound wave, and is also equipped with a broadband FFT analyzer. The scientific novelty of the research lies in the consideration of composer folklorism from the point of view of the timbre acoustic model of authentic singing, and the adaptation of expressive means is investigated not within the framework of the style and manner of the melodic presentation of the material of individual composers, but from the point of view of the timbre content of the folklore voice. The article shows how the influence of ethno-singing on composer’s creativity is revealed through the prism of authentic timbre and the given spectral examples.

A Novel Heterogeneous Parallel Convolution Bi-LSTM for Speech Emotion Recognition

Speech emotion recognition is a substantial component of natural language processing (NLP). It has strict requirements for the effectiveness of feature extraction and that of the acoustic model. With that in mind, a Heterogeneous Parallel Convolution Bi-LSTM model is proposed to address the challenges. It consists of two heterogeneous branches: the left one contains two dense layers and a Bi-LSTM layer, while the right one contains a dense layer, a convolution layer, and a Bi-LSTM layer. It can exploit the spatiotemporal information more effectively, and achieves 84.65%, 79.67%, and 56.50% unweighted average recalls on the benchmark databases EMODB, CASIA, and SAVEE, respectively. Compared with the previous research results, the proposed model achieves better performance stably.