Data Driven Neural Speech Enhancement for Smart Healthcare in Consumer Electronics Applications

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F60076658%3A12310%2F24%3A43908873" target="_blank" >RIV/60076658:12310/24:43908873 - isvavai.cz</a>

Výsledek na webu

<a href="https://ieeexplore.ieee.org/document/10496958" target="_blank" >https://ieeexplore.ieee.org/document/10496958</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1109/TCE.2024.3387740" target="_blank" >10.1109/TCE.2024.3387740</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Data Driven Neural Speech Enhancement for Smart Healthcare in Consumer Electronics Applications

Popis výsledku v původním jazyce

This paper presents the practical response and performance-aware development of online speech enhancement from a consumer electronic perspective. To improve the efficiency of human-machine interaction, speech can play a vital role as a transmission medium on the Internet of Medical Things (IoM). However, some intelligent speech recognition systems cannot preserve the confidentiality of speech data. Additionally, the preservation of privacy is onerous, especially for model training and speech recognition in real-time. The recent development of big data-oriented wireless technologies associated with edge computing, interconnected devices of the Internet of Medical Things (IoMT), and big data analytics has great demand for connected human-machine interaction for many applications like automated cars, health monitoring, and consumer personal health care monitoring systems. Although big data-oriented wireless technologies serve these applications, the challenge remains of ignoring emotional care. This paper starts by explaining how to make a neural network-based architecture that can improve the speech of multichannel first-order Ambisonics mixtures and lower the need for human intervention through ambient intelligence (AmI). This will make the system work better overall in medical situations. Second, we demonstrate the effectiveness of different noise estimation techniques on proposed modulation domain processing (MDP) applications in smart hospitals, including electronic medical documentation, disease diagnosis, and evaluation. The proposed approach outperforms the enhancement of the conventional modulation domain in the cortex with several objective evaluation parameters such as Log Likelihood Ratio (LLR), Weighted Spectral Slope (WSS), Perceptual Evaluation of Speech Quality (PESQ), Csig and segmental (SNR seg.) Different noise estimators are used to figure out what effect the system has on different spectral modification parameters, like the over-subtraction factor and the modification domain. The experimental results show that the MDP system achieves better performance in terms of SNRseg. scores (49%) for the state-of-the-art consumer electronics perspective in a health care system. The proposed framework would greatly contribute to personalized communication health monitoring by consumers in a noisy environment.

Název v anglickém jazyce

Data Driven Neural Speech Enhancement for Smart Healthcare in Consumer Electronics Applications

Popis výsledku anglicky

This paper presents the practical response and performance-aware development of online speech enhancement from a consumer electronic perspective. To improve the efficiency of human-machine interaction, speech can play a vital role as a transmission medium on the Internet of Medical Things (IoM). However, some intelligent speech recognition systems cannot preserve the confidentiality of speech data. Additionally, the preservation of privacy is onerous, especially for model training and speech recognition in real-time. The recent development of big data-oriented wireless technologies associated with edge computing, interconnected devices of the Internet of Medical Things (IoMT), and big data analytics has great demand for connected human-machine interaction for many applications like automated cars, health monitoring, and consumer personal health care monitoring systems. Although big data-oriented wireless technologies serve these applications, the challenge remains of ignoring emotional care. This paper starts by explaining how to make a neural network-based architecture that can improve the speech of multichannel first-order Ambisonics mixtures and lower the need for human intervention through ambient intelligence (AmI). This will make the system work better overall in medical situations. Second, we demonstrate the effectiveness of different noise estimation techniques on proposed modulation domain processing (MDP) applications in smart hospitals, including electronic medical documentation, disease diagnosis, and evaluation. The proposed approach outperforms the enhancement of the conventional modulation domain in the cortex with several objective evaluation parameters such as Log Likelihood Ratio (LLR), Weighted Spectral Slope (WSS), Perceptual Evaluation of Speech Quality (PESQ), Csig and segmental (SNR seg.) Different noise estimators are used to figure out what effect the system has on different spectral modification parameters, like the over-subtraction factor and the modification domain. The experimental results show that the MDP system achieves better performance in terms of SNRseg. scores (49%) for the state-of-the-art consumer electronics perspective in a health care system. The proposed framework would greatly contribute to personalized communication health monitoring by consumers in a noisy environment.

Druh

J_imp - Článek v periodiku v databázi Web of Science

CEP obor

—

OECD FORD obor

20201 - Electrical and electronic engineering

Projekt

—

Návaznosti

I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Rok uplatnění

2024

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název periodika

IEEE TRANSACTIONS ON CONSUMER ELECTRONICS

ISSN

0098-3063

e-ISSN

1558-4127

Svazek periodika

70

Číslo periodika v rámci svazku

2

Stát vydavatele periodika

US - Spojené státy americké

Počet stran výsledku

11

Strana od-do

4828-4838

Kód UT WoS článku

001303863100006

EID výsledku v databázi Scopus

2-s2.0-85190337273