A computational model for auditory saliency of environmental sound.

The purpose of this study was to identify the main research themes and knowledge structures in the field of urban soundscape. With the continuous expansion of research work in the field of urban soundscape, it has become necessary to carry out a systematic analysis. CiteSpace was used to conduct an information visualization analysis of high-quality literature related to urban soundscape research in the WoS database from 1976 to 2021. The results revealed the following: (1) In terms of research content, research hotspots center on noise, perception, and quality, while focusing on theory and methodology. (2) In terms of research methods, the Perceptual Restorative Soundscape Scale has gradually become the main method of soundscape research. With the development of sound acquisition technology and sound simulation technology, the soundscape perception model will undergo an iterative process of updating. (3) In terms of research objects, most of the research focuses on the soundscape of outdoor environments (such as urban parks, tourist attractions, and historical blocks) together with the influences and preferences for different types of soundscapes. The research results can provide reference for research and planning as well as the design practice of urban soundscape.

People in urban areas frequently use parks for recreation and outdoor activities. Owing to the complexity of the outdoor environment, there have only been a few attempts to understand the effect of the thermal environment on people's use of outdoor spaces. This paper therefore seeks to determine the relationship between the thermal environment, park use and behavioural patterns in an urban area of Sweden. The methods used include structured interviews, unobtrusive observations of the naturally occurring behaviour and simultaneous measurements of thermal comfort variables, i.e., air temperature, air humidity, wind speed and global radiation. The thermal environment is investigated through the mean radiant temperature (Tmrt) and the predicted mean vote (PMV) index. The outcome is compared to the subjective behaviour and thermal sensation of the interviewees. It is found that the thermal environment, access and design are important factors in the use of the park. In order to continue to use the park when the thermal conditions become too cold or too hot for comfort, people improve their comfort conditions by modifying their clothing and by choosing the most supportive thermal opportunities available within the place. The study also shows that psychological aspects such as time of exposure, expectations, experience and perceived control may influence the subjective assessment. Comparison between the thermal sensation of the interviewees and the thermal sensation assessed by the PMV index indicates that steady-state models such as the PMV index may not be appropriate for the assessment of short-term outdoor thermal comfort, mainly because they are unable to analyse transient exposure.

Learning data representations directly from the data itself is an approach that has shown great success in different pattern recognition problems, outperforming state-of-the-art feature extraction schemes for different tasks in computer vision, speech recognition and natural language processing. Representation learning applies unsupervised and supervised machine learning methods to large amounts of data to find building-blocks that better represent the information in it. Digitized histopathology images represents a very good testbed for representation learning since it involves large amounts of high complex, visual data. This paper presents a comparative evaluation of different supervised and unsupervised representation learning architectures to specifically address open questions on what type of learning architectures (deep or shallow), type of learning (unsupervised or supervised) is optimal. In this paper we limit ourselves to addressing these questions in the context of distinguishing between anaplastic and non-anaplastic medulloblastomas from routine haematoxylin and eosin stained images. The unsupervised approaches evaluated were sparse autoencoders and topographic reconstruct independent component analysis, and the supervised approach was convolutional neural networks. Experimental results show that shallow architectures with more neurons are better than deeper architectures without taking into account local space invariances and that topographic constraints provide useful invariant features in scale and rotations for efficient tumor differentiation.

Many successful models for predicting attention in a scene involve three main steps: convolution with a set of filters, a center-surround mechanism and spatial pooling to construct a saliency map. However, integrating spatial information and justifying the choice of various parameter values remain open problems. In this paper we show that an efficient model of color appearance in human vision, which contains a principled selection of parameters as well as an innate spatial pooling mechanism, can be generalized to obtain a saliency model that outperforms state-of-the-art models. Scale integration is achieved by an inverse wavelet transform over the set of scale-weighted center-surround responses. The scale-weighting function (termed ECSF) has been optimized to better replicate psychophysical data on color appearance, and the appropriate sizes of the center-surround inhibition windows have been determined by training a Gaussian Mixture Model on eye-fixation data, thus avoiding ad-hoc parameter selection. Additionally, we conclude that the extension of a color appearance model to saliency estimation adds to the evidence for a common low-level visual front-end for different visual tasks.

In our paper we examine the usage prosodic features in speech recognition, with a special attention payed to agglutinating and fixed stress languages. The used prosodic features, acoustic-prosodic pre-processing, and segmentation in terms of prosodic units are presented in details. We use the expression "prosodic unit" in order to make a difference from prosodic phrases, which are longer. We trained a HMM-based prosodic segmenter reliing on fundamental frequency and intensity of speech. The output of the prosodic segmenter is used for N-best lattice rescoring in parallel with a simplified bigram language model in a continuous speech recognizer, in order to improve speech recognition performance. Experiments for Hungarian language show aWER reduction of about 4% using a simple lattice rescoring.

In this chapter we examine the usage of prosodic features in speech recognition, with a special attention payed to agglutinating and fixed stress languages. Current knowledge in speech prosody exploitation is addressed in the introduction. The used prosodic features, acousticprosodic pre-processing, and segmentation in terms of prosodic units are presented in details. We use the expression "prosodic unit" in order to differentiate them from prosodic phrases, which are usually longer. We trained a HMM-based prosodic segmenter relying on fundamental frequency and intensity of speech. The output of this prosodic segmenter is used for N-best lattice rescoring in parallel with a simplified bigram language model in a continuous speech recognizer, in order to improve speech recognition performance. Experiments for Hungarian language show a WER reduction of about 4% using a simple lattice rescoring. The performance of the prosodic segmenter is also investigated in comparison with our earlier experiments.

As an acute respiratory disease, scarlet fever has great harm to public health. Some evidence indicates that the time distribution pattern of heavy PM2.5 pollution occurrence may have an impact on health risks. This study aims to reveal the relation between scaling features in high-concentrations PM2.5 (HC-PM2.5) evolution and scarlet fever incidence (SFI). Based on the data of Hong Kong from 2012 to 2019, fractal box-counting dimension (D) is introduced to capture the scaling features of HC-PM2.5. It has been found that index D can quantify the time distribution of HC-PM2.5, and lower D values indicate more cluster distribution of HC-PM2.5. Moreover, scale-invariance in HC-PM2.5 at different time scales has been discovered, which indicates that HC-PM2.5 occurrence is not random but follows a typical power-law distribution. Next, the exposure-response relationship between SFI and scale-invariance in HC-PM2.5 is explored by Distributed lag non-linear model, in conjunction with meteorological factors. It has been discovered that scale-invariance in HC-PM2.5 has a nonlinear effect on SFI. Low and moderate D values of HC-PM2.5 are identified as risk factors for SFI at small time-scale. Moreover, relative risk shows a decreasing trend with the increase of exposure time. These results suggest that exposure to short-term clustered HC-PM2.5 makes individual more prone to SFI than exposure to long-term uniform HC-PM2.5. This means that individuals in slightly-polluted regions may face a greater risk of SFI, once the PM2.5 concentration keeps rising. In the future, it is expected that the relative risk of scarlet fever for a specific region can be estimated based on the quantitative analysis of scaling features in high-concentrations PM2.5 evolution.

In order to elucidate the dynamics of cellular processes that are induced in context with intercellular communication, defined events along the signal transduction cascade and subsequent activation steps have to be analyzed on the level of individual cells and correlated with each other. Here we present an approach that allows the initiation of cell-cell or cell-particle interactions and the analysis of cellular reactions within various regimes while the identity of each individual cell is preserved. It utilizes dielectrophoresis (DEP) and microfluidics in a lab-on-chip system. With high spatial and temporal precision we contacted single T cells with functionalized microbeads and monitored their immediate cytosolic Ca(2+) response. After this, the cells were released from the chip system and cultivated further. Expression of the activation marker molecule CD69 was analyzed the next day and correlated with the previously recorded Ca(2+) signal for each individual cell. We found a significant difference in the patterns of Ca(2+) traces between activated and non-activated cells, which shows that Ca(2+) signals in T cells can provide early information about a later reaction of the cell. Although T cells are non-excitable cells, we also observed irregular Ca(2+) transients upon exposure to the DEP field only. These Ca(2+) signals depended on exposure time, electric field strength and field frequency. By minimizing their occurrence rate, we could identify experimental conditions that caused the least interference with the physiology of the cell.

Both the timing (i.e., when) and amount (i.e., how much) of language exposure affect language-learning outcomes. We compared speech recognition accuracy across three listener groups for whom the order (first versus second) and dominance (dominant versus non-dominant) of two languages, English and Spanish, varied: one group of Spanish heritage speakers (SHS; L2-English dominant; L1-Spanish non-dominant) and two groups of late onset L2 learners (L1-dominant English/Spanish learners and L1-dominant Spanish/English learners). Sentence-final word recognition accuracy in both English and Spanish was assessed across three “easy” versus “difficult” listening conditions: (1) signal-to-noise ratio (SNR; +5 dB SNR versus 0 dB SNR), (2) sentence predictability (high versus low sentence predictability), and (3) speech style (clear versus plain speech style). Overall, SHS English recognition accuracy was equivalent to that of the L1-dominant English Spanish learners, whereas SHS Spanish recognition accuracy was substantially lower than that of the L1-dominant Spanish English learners. Moreover, while SHS benefitted in both languages from the “easy” listening conditions, they were more adversely affected by (i.e., they recognized fewer words) the presence of higher noise and lower predictability in their non-dominant L1 Spanish compared to their dominant L2 English. These results identify both a benefit and limit on the influence of early exposure. Specifically, the L2-dominant heritage speakers displayed L1-like speech recognition in their dominant-L2, as well as generally better recognition in their non-dominant L1 than late onset L2 learners. Yet, subtle recognition accuracy differences between SHS and L1-dominant listeners emerged under relatively difficult communicative conditions.

Cone heater ignition tests of liquid fuels

Plasma-enhanced atomic layer deposition (PE-ALD) is widely used for dielectric deposition in semiconductor fabrication due to its ability to operate at low temperatures while having high precision control. The PE-ALD process consists of two subcycles: precursor dosing and plasma exposure with gas purging and filling in between. In the PE-ALD of SiO2, a Si-containing precursor is first deposited on the surface, usually in a plasma-free environment. The surface is then exposed to an oxygen-containing plasma during which the residual components of the precursor are removed and the Si oxidized. Various factors affect the outcome of SiO2 PE-ALD, such as exposure times during each step, steric hindrance of the Si precursor, and plasma properties, such as the energy of ions incident onto the film. The results from computational investigations of the first layers of SiO2 PE-ALD at both reactor (cm) and feature (nm) scales are discussed in this paper. The example system uses bis(tertiary-butylamino)silane, SiH2[NH(C4H9)]2 as the silicon precursor during dosing and plasmas operating in Ar/O2 gas mixtures during the oxidation step. Parametric studies were performed for blanket deposition, as well as deposition in trenches and vias while varying power, pressure, plasma exposure time, aspect ratio, and ligand retention in the film. The general trends show that conditions that reduce the fluence of reactive oxygen species typically decrease the O/Si ratio, increase the vacancies in the films, and decrease the order of the film. Conditions that result in higher ion fluxes having higher energies produce the same result due to sputtering. The retention of ligand groups from the precursor significantly decreased growth rates while increasing vacancies and reducing the O/Si ratio.

A new, simple and practical way of fusing audio and visual information to enhance audiovisual automatic speech recognition within the framework of an application of large-vocabulary speech recognition of French Canadian speech is presented, and the experimental methodology is described in detail. The visual information about mouth shape is extracted off-line using a cascade of weak classifiers and a Kalman filter, and is combined with the large-vocabulary speech recognition system of the Centre de Recherche Informatique de Montreal. The visual classification is performed by a pair-wise kernel-based linear discriminant analysis (KLDA) applied on a principal component analysis (PCA) subspace, followed by a binary combination and voting algorithm on 35 French phonetic classes. Three fusion approaches are compared: (1) standard low-level feature-based fusion, (2) decision-based fusion within the framework of the transferable belief model (an interpretation of the Dempster-Shafer evidential theory), and (3) a combination of (1) and (2). For decision-based fusion, the audio information is considered to be a precise Bayesian source, while the visual information is considered an imprecise evidential source. This treatment ensures that the visual information does not significantly degrade the audio information in situations where the audio performs well (e.g., a controlled noise-free environment). Results show significant improvement in the word error rate to a level comparable to that of more sophisticated systems. To the authors' knowledge, this work is the first to address large-vocabulary audiovisual recognition of French Canadian speech and decision-based audiovisual fusion within the transferable belief model.

Artificial intelligence has revolutionized image and speech recognition, but the neural network fitting method has limitations. Neuromorphic chips that mimic biological neurons can better simulate the brain’s information processing mechanism. As the basic computing component of the new neuromorphic network, the new neural computing unit’s design and implementation have important significance; however, complex dynamical features come with a high computational cost: approximate computing has unique advantages, in terms of optimizing the computational cost of neural networks, which can solve this problem. This paper proposes a hardware implementation of an approximate spiking neuron structure, based on a simplified piecewise linear model (SPWL), to optimize power consumption and area. The proposed structure can achieve five major neuron spiking generation patterns. The proposed design was synthesized and compared to similar designs, to evaluate its potential advantages and limitations. The results showed that the approximate spiking neuron had the lowest computational cost and the fastest computation speed. A typical spiking neural network was constructed, to test the usability of the SPWL model. The results showed that the proposed approximate spiking neuron could work normally in the spiking neural network, and achieved an accuracy of 94% on the MNIST dataset.

The paper presents an implemented model for priming speech recognition, using contextual information about salient entities. The underlying hypothesis is that, in human-robot interaction, speech recognition performance can be improved by exploiting knowledge about the immediate physical situation and the dialogue history. To this end, visual salience (objects perceived in the physical scene) and linguistic salience (objects, events already mentioned in the dialogue) are integrated into a single cross-modal salience model. The model is dynamically updated as the environment changes. It is used to establish expectations about which words are most likely to be heard in the given context. The update is realised by continuously adapting the word-class probabilities specified in a statistical language model. The paper discusses the motivations behind the approach, and presents the implementation as part of a cognitive architecture for mobile robots. Evaluation results on a test suite show a statistically significant improvement of salience-driven priming speech recognition (WER) over a commercial baseline system.

The mel-frequency cepstral coefficient (MFCC) or perceptual linear prediction (PLP) feature extraction typically used for automatic speech recognition (ASR) employ several principles which have known counterparts in the cochlea and auditory nerve: frequency decomposition, mel- or bark-warping of the frequency axis, and compression of amplitudes. It seems natural to ask if one can profitably employ a counterpart of the next physiological processing step, synaptic adaptation. We, therefore, incorporated a simplified model of short-term adaptation into MFCC feature extraction. We evaluated the resulting ASR performance on the AURORA 2 and AURORA 3 tasks, in comparison to ordinary MFCCs, MFCCs processed by RASTA, and MFCCs processed by cepstral mean subtraction (CMS), and both in comparison to and in combination with Wiener filtering. The results suggest that our approach offers a simple, causal robustness strategy which is competitive with RASTA, CMS, and Wiener filtering and performs well in combination with Wiener filtering. Compared to the structurally related RASTA, our adaptation model provides superior performance on AURORA 2 and, if Wiener filtering is used prior to both approaches, on AURORA 3 as well.