Stream fusion for multi-stream automatic speech recognition

Multi-stream automatic speech recognition (MS-ASR) has been confirmed to boost the recognition performance in noisy conditions. In this system, the generation and the fusion of the streams are the essential parts and need to be designed in such a way to reduce the effect of noise on the final decision. This paper shows how to improve the performance of the MS-ASR by targeting two questions; (1) How many streams are to be combined, and (2) how to combine them. First, we propose a novel approach based on stream reliability to select the number of streams to be fused. Second, a fusion method based on Parallel Hidden Markov Models is introduced. Applying the method on two datasets (TIMIT and RATS) with different noises, we show an improvement of MS-ASR.     

Affective speaker state analysis in the presence of reverberation

Little is known about the impact of reverberation on performance of running speaker state classification systems. This study thus aims to approach the topic by measuring effects on a state-of-the-art engine with consideration of six public room impulse responses for convolution of the affective speech signals of three standard datasets comprising of emotion and interest. Speech data thereby is given by this year’s INTERSPEECH Paralinguistic Challenge corpus TUM AVIC and the frequently used Berlin and eNTERFACE sets. The room impulse responses comprise rooms in private apartments, chapels, a factory hall, and a van. Speaker independent performance after speaker adaptation is investigated. To cope with reverberation, matched condition learning and acoustic space adaptation are considered as efficient means. By that a report is provided on suitability of feature types given the type of impulse response. In the result almost all occurring corruption arising from reverberation can be restored, yet the general impact varies with the type of room or acoustic environment.     

Deep Scalogram Representations for Acoustic Scene Classification

Spectrogram representations of acoustic scenes have achieved competitive performance for acoustic scene classification. Yet, the spectrogram alone does not take into account a substantial amount of time-frequency information. In this study, we present an approach for exploring the benefits of deep scalogram representations, extracted in segments from an audio stream. The approach presented firstly transforms the segmented acoustic scenes into bump and morse scalograms, as well as spectrograms; secondly, the spectrograms or scalograms are sent into pre-trained convolutional neural networks; thirdly, the features extracted from a subsequent fully connected layer are fed into (bidirectional) gated recurrent neural networks, which are followed by a single highway layer and a softmax layer; finally, predictions from these three systems are fused by a margin sampling value strategy. We then evaluate the proposed approach using the acoustic scene classification data set of 2017 IEEE AASP Challenge on Detection and Classification of Acoustic Scenes and Events (DCASE). On the evaluation set, an accuracy of 64.0% from bidirectional gated recurrent neural networks is obtained when fusing the spectrogram and the bump scalogram, which is an improvement on the 61.0% baseline result provided by the DCASE 2017 organisers. This result shows that extracted bump scalograms are capable of improving the classification accuracy, when fusing with a spectrogram-based system.      

133Cs and 23Na MAS NMR Spectroscopy of Molybdate Crystallization in Model Nuclear Glasses

Sodium borosilicate base glasses modeled on French nuclear waste materials were prepared to test the dependence of crystallization product quantity and distribution on cesium‐ and molybdenum‐loading and glass cooling rate. Scanning electron microscopy shows the presence of micrometer‐sized domains of Mo‐rich crystalline precipitates. X‐ray diffraction patterns are complex but reveal the presence of sodium molybdates and CsNaMoO₄·2H₂O. ¹³³Cs and ²³Na magic‐angle spinning NMR spectroscopy exhibit distinct peaks for glassy and crystalline phases which can be quantified to obtain the identities of the individual compounds that are formed as well as the fractions of these nuclei in particular crystalline phases. In these model systems, 1 mol% Mo can be entirely incorporated into the glassy network whereas 2.5 and 5 mol% Mo produce significant quantities of crystalline phases, with little dependence on cooling rate. Cesium content appears to have a weak influence on crystallization behavior. Sodium molybdate and sodium‐cesium molybdate hydrate are the dominant devitrification phases in all cases.     

Whodunnit – Searching for the most important feature types signalling emotion-related user states in speech

In this article, we describe and interpret a set of acoustic and linguistic features that characterise emotional/emotion-related user states – confined to the one database processed: four classes in a German corpus of children interacting with a pet robot. To this end, we collected a very large feature vector consisting of more than 4000 features extracted at different sites. We performed extensive feature selection (Sequential Forward Floating Search) for seven acoustic and four linguistic types of features, ending up in a small number of ‘most important’ features which we try to interpret by discussing the impact of different feature and extraction types. We establish different measures of impact and discuss the mutual influence of acoustics and linguistics.     

Electron Transfer Reactions in Pulping Systems (III): A Study of Steric Effects in Lignin Model/Aq Reactions

Abstract

The reactions of five B-aryl ether lignin model dimers with three anthrahydroquinone (AHQ) analogs have been studied. Some of the models and AHQ analogs have bulky substituents strategically located in positions which would possibly inhibit adduct reactions but not single electron transfer (SET) reactions. The fact that the model fragmentation efficiencies were the same for both steri-cally hindered and unhindered AHQ analogs indicates that the reaction mechanism cannot involve a rate determining adduct formation step. The results can be best explained either by an SET mechanism or a mechanism which involves quinoneraethide generation as a slow step, followed by adduct, SET, or other steps. Placing methyl groups on the S-carbon of the models favored model fragmentation reactions by NaOH. The B-methyl group may be promoting fragmentation reaction rates and/or retarding the rates of competing side reactions, such as vinyl ether generation.