東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁 到查詢結果 [ null ]

切換: 標籤 | MARC模式 | ISBD

Towards Context-Aware and Trustworth...

Zhang, Shaohu,

FindBook

Google Book

Amazon

博客來

Towards Context-Aware and Trustworthy Voice Assistants /

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Towards Context-Aware and Trustworthy Voice Assistants // Shaohu Zhang.
作者:	Zhang, Shaohu,
面頁冊數:	1 electronic resource (150 pages)
附註:	Source: Dissertations Abstracts International, Volume: 85-04, Section: B.
Contained By:	Dissertations Abstracts International85-04B.
標題:	Sensitivity analysis. -
電子資源:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30615806
ISBN:	9798380483070

Towards Context-Aware and Trustworthy Voice Assistants /
Zhang, Shaohu,

Towards Context-Aware and Trustworthy Voice Assistants /Shaohu Zhang. - 1 electronic resource (150 pages)

Source: Dissertations Abstracts International, Volume: 85-04, Section: B.

The use of voice-control technology has been pervasive in laptops, smartphones, home voice assistants, and Internet of Things (IoT) devices. While voice assistants provide convenience through automation and control of home appliances, the lack of context awareness of the voice assistants (VA) comes with security and privacy risks. As a result, VAs have been shown to be vulnerable to misactivation, replay attacks, and inference of physical attributes. This dissertation investigates the potential of acoustic sensing and signal modification to integrate human physical context (such as hand gesture and voice attributes) into VAs, thus making them more context-aware and reliable.We first introduce the concept of using a gesture-based authentication system called HandLock, which uses built-in microphones and speakers of VA to generate and sense inaudible acoustic signals to detect the presence of a known hand gesture for authentication. The proposed approach can act as a second-factor authentication (2-FA) for performing specific sensitive operations like confirming online purchases through VAs.The second problem is that VA is susceptible to misactivation due to either suboptimal wake-word recognition engine or artificial manipulation of a wake word's pronunciation, creating both security and privacy concerns. Existing privacy controls for VAs are neither effective nor user-friendly. Inspired by recent studies showing that visual gaze plays an important role in interacting with VAs, we propose a device-free and non-obtrusive acoustic sensing system called HeadTalk, to thwart the misactivation of VAs. The proposed technique can accurately infer a speaker's head orientation, allowing VAs to record and transmit audio data only when they detect a human speaker facing them from a distance.The third problem is the opportunity for sensitive data harvesting from recorded audio. Numerous privacy-preserving measures have been proposed for hiding the speaker's identity while maintaining speech intelligibility. However, existing works do not consider the overall tradeoff between speech utility, speaker verification, and inference of voice attributes, including emotional state, age, accent, and gender. In this study, we propose VoicePM, a new tradeoff metric to capture voice biometrics as well as different voice attributes. We propose a novel voice privacy measurement framework, to study the feasibility of applying different state-of-the-art voice anonymization solutions to achieve the optimum tradeoff between privacy and utility.This dissertation also presents detailed technical development, implementation, and extensive experiments of these three systems.

English

ISBN: 9798380483070Subjects--Topical Terms:

3560752
Sensitivity analysis.

Towards Context-Aware and Trustworthy Voice Assistants /
LDR:04025nmm a22003853i 4500 001 2430331
005 20250519115943.5
006 m o d
007 cr|nu||||||||
008 261011s2023 miu||||||m |||||||eng d
020 $a 9798380483070
035 $a (MiAaPQD)AAI30615806
035 $a (MiAaPQD)NCState_Univ18402041022
035 $a AAI30615806
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Zhang, Shaohu, $e author. $3 3818597
245 1 0 $a Towards Context-Aware and Trustworthy Voice Assistants / $c Shaohu Zhang.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2023
300 $a 1 electronic resource (150 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 85-04, Section: B.
500 $a Advisors: Shahzad, Muhammad; Das, Anupam Committee members: Wong, Chau-Wai; Enck, William.
502 $b Ph.D. $c North Carolina State University $d 2023.
520 $a The use of voice-control technology has been pervasive in laptops, smartphones, home voice assistants, and Internet of Things (IoT) devices. While voice assistants provide convenience through automation and control of home appliances, the lack of context awareness of the voice assistants (VA) comes with security and privacy risks. As a result, VAs have been shown to be vulnerable to misactivation, replay attacks, and inference of physical attributes. This dissertation investigates the potential of acoustic sensing and signal modification to integrate human physical context (such as hand gesture and voice attributes) into VAs, thus making them more context-aware and reliable.We first introduce the concept of using a gesture-based authentication system called HandLock, which uses built-in microphones and speakers of VA to generate and sense inaudible acoustic signals to detect the presence of a known hand gesture for authentication. The proposed approach can act as a second-factor authentication (2-FA) for performing specific sensitive operations like confirming online purchases through VAs.The second problem is that VA is susceptible to misactivation due to either suboptimal wake-word recognition engine or artificial manipulation of a wake word's pronunciation, creating both security and privacy concerns. Existing privacy controls for VAs are neither effective nor user-friendly. Inspired by recent studies showing that visual gaze plays an important role in interacting with VAs, we propose a device-free and non-obtrusive acoustic sensing system called HeadTalk, to thwart the misactivation of VAs. The proposed technique can accurately infer a speaker's head orientation, allowing VAs to record and transmit audio data only when they detect a human speaker facing them from a distance.The third problem is the opportunity for sensitive data harvesting from recorded audio. Numerous privacy-preserving measures have been proposed for hiding the speaker's identity while maintaining speech intelligibility. However, existing works do not consider the overall tradeoff between speech utility, speaker verification, and inference of voice attributes, including emotional state, age, accent, and gender. In this study, we propose VoicePM, a new tradeoff metric to capture voice biometrics as well as different voice attributes. We propose a novel voice privacy measurement framework, to study the feasibility of applying different state-of-the-art voice anonymization solutions to achieve the optimum tradeoff between privacy and utility.This dissertation also presents detailed technical development, implementation, and extensive experiments of these three systems.
546 $a English
590 $a School code: 0155
650 4 $a Sensitivity analysis. $3 3560752
650 4 $a Privacy. $3 528582
650 4 $a Systems design. $3 3433840
650 4 $a Signal processing. $3 533904
650 4 $a Design. $3 518875
650 4 $a Electrical engineering. $3 649834
690 $a 0389
690 $a 0544
710 2 $a North Carolina State University. $e degree granting institution. $3 3765716
720 1 $a Shahzad, Muhammad $e degree supervisor.
720 1 $a Das, Anupam $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 85-04B.
790 $a 0155
791 $a Ph.D.
792 $a 2023
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30615806