A5011297325- Speech and Audio Processing
- Speech Recognition and Synthesis
- Advanced Adaptive Filtering Techniques
- Voice and Speech Disorders
- Hearing Loss and Rehabilitation
- Blind Source Separation Techniques
- Music and Audio Processing
- Phonetics and Phonology Research
- Indoor and Outdoor Localization Technologies
- Gut microbiota and health
- Underwater Acoustics Research
- Osteoarthritis Treatment and Mechanisms
- Protein Hydrolysis and Bioactive Peptides
- Probiotics and Fermented Foods
- Diet and metabolism studies
- Meat and Animal Product Quality
- Advanced Data Compression Techniques
- Water Quality Monitoring Technologies
- Water Quality Monitoring and Analysis
- Knee injuries and reconstruction techniques
- Oral microbiology and periodontitis research
- Wireless Signal Modulation Classification
- Direction-of-Arrival Estimation Techniques
- Metabolomics and Mass Spectrometry Studies
- Digestive system and related health
Alibaba Group (China)
2023-2024
China University of Mining and Technology
2024
Ningbo Science and Technology Bureau
2022
University of Sheffield
2019-2020
Carl von Ossietzky Universität Oldenburg
2015-2018
Hearing4all
2015-2018
Fraunhofer Institute for Digital Media Technology
2010-2017
Fraunhofer Society
2013
Improving the accuracy of personalised speech recognition for speakers with dysarthria is a challenging research field. In this paper, we explore an approach that non-linearly modifies tempo to reduce mismatch between typical and atypical speech. Speech analysis at phonetic level accomplished using forced-alignment process from traditional GMM-HMM in automatic (ASR). Estimated adjustments are applied directly acoustic features rather than time-domain signals. Two approaches considered: i)...
This paper presents an improved transfer learning framework applied to robust personalised speech recognition models for speakers with dysarthria. As the baseline of learning, a state-of-the-art CNN-TDNN-F ASR acoustic model trained solely on source domain data is adapted onto target via neural network weight adaptation limited available from dysarthric speakers. Results show that linear weights in layers play most important role modelling evaluated using UASpeech corpus, achieving averaged...
Bifidobacterium pseudolongum is widely exists in mammal gut and its abundance associated with human animal health. The present study aimed to investigate the potential mechanisms of B. CCFM1253 on protecting against lipopolysaccharide (LPS)-induced acute liver injury (ALI) by metagenomic analysis metabolomic profiles.Bifidobacterium preintervention remarkably attenuated influence LPS serum alanine transaminase aspartate amino transferase activities. inflammation responses (tumor necrosis...
The reverberation time (RT) and the early-to-late ratio (ELR) are two key parameters commonly used to characterize acoustic room environments. In contrast conventional blind estimation methods that process separately, we propose a model for joint predict RT ELR simultaneously from single-channel speech signals either full-band or sub-band frequency data, which is referred as parameter estimator (jROPE). An artificial neural network employed learn mapping observations classes....
This paper presents extended techniques aiming at the improvement of automatic speech recognition (ASR) in single-channel scenarios context REVERB (REverberant Voice Enhancement and Recognition Benchmark) challenge. The focus is laid on development analysis ASR front-end technologies covering enhancement feature extraction. Speech performed using a joint noise reduction dereverberation system spectral domain based estimates late reverberation power densities (PSDs). To obtain reliable...
Blind estimation of acoustic room parameters such as the reverberation time $T_\mathrm{60}$ and direct-to-reverberation ratio ($\mathrm{DRR}$) is still a challenging task, especially in case blind from reverberant speech signals. In this work, novel approach proposed for joint $\mathrm{DRR}$ wideband noisy conditions. 2D Gabor filters arranged filterbank are exploited extracting features, which then used input to multi-layer perceptron (MLP). The MLP output neurons correspond specific pairs...
A novel method for blind estimation of the reverberation time (RT60) is proposed based on applying spectro-temporal modulation filters to time-frequency representations. 2D-Gabor arranged in a filterbank enable an analysis properties temporal, spectral, and filtering this task. Features are used as input multi-layer perceptron (MLP) classifier combined with simple decision rule that attributes specific RT60 given utterance allows assess reliability approach different resolutions...
There has been much recent interest in building continuous speech recognition systems for people with severe impairments, e.g., dysarthria. However, the datasets that are commonly used typically designed tasks other than ASR development, or they contain only isolated words. As such, overlap prompts read by speakers. Previous evaluations have often neglected this, using language models (LMs) trained on non-disjoint training and test data, potentially producing unrealistically optimistic...
Room acoustic parameters that characterize environments can help to improve signal enhancement algorithms such as for dereverberation, or automatic speech recognition by adapting models the current parameter set. The reverberation time (RT) and early-to-late ratio (ELR) are two key parameters. In this paper, we propose a blind ROom Parameter Estimator (ROPE) based on an artificial neural network learns mapping discrete ranges of RT ELR from single-microphone signals. Auditory-inspired...
This work analyzes the influence of reverberation on automatic speech recognition (ASR) systems and how to compensate its influence, with special focus important acoustical parameters i.e. room time T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">60</sub> clarity index C xmlns:xlink="http://www.w3.org/1999/xlink">50</sub> . A multilayer perceptron (MLP) using features a spectro-temporal filter bank as input is employed identify acoustic...
We present a lightweight neural network with attentive score loss for frame-wise personalized voice activity detection (i.e., AS-pVAD). Instead of using an external speaker embedding extractor large number parameters, AS-pVAD employs internal model to extract the target embedding. A novel constraint is proposed better exploit such clues pVAD compared conventional concatenation. Through joint training regular VAD, can be further improved identify in enrollment cases while it able function as...
Speech communication is the most natural form of human interaction. Communication by means telephones, mobile phones or video-conference systems common nowadays especially amongst younger persons. In past years, also a growing amount elderly people has started to extensively use since more and live apart from their relatives, friends acquaintances. However, suffer hearing loss, which often prevents them using acoustic devices. While approximately every second European adult age 65+ loss that...
Short peptides have gained widespread utilization as functional constituents in the development of foods due to their remarkable biological activity. Previous investigations established positive influence oysters on testosterone biosynthesis, although underlying mechanism remains elusive. This study aims assess impact three dipeptides derived from oxidative stress state TM3 cells induced by AAPH while concurrently examining alterations cellular biosynthesis capacity. The investigation...
This paper investigates four single-channel speech dereverberation algorithms, i.e., two unsupervised approaches based on (i) spectral enhancement and (ii) linear prediction, as well supervised relying machine learning which incorporate deep neural networks to predict either (iii) the magnitude spectrogram or (iv) ideal ratio mask. The relative merits of algorithms in terms several objective measures, automatic recognition performance, robustness against noise, variations between simulated...