Speech security on human-computer interaction
Grant-in-Aid for Research Activity Start-up
1002:Human informatics, applied informatics and related fields
|Japan Advanced Institute of Science and Technology
MAWALIM CandyOlivia 北陸先端科学技術大学院大学, 先端科学技術研究科, 助教 (10963720)
|Project Period (FY)
2022-08-31 – 2024-03-31
Granted (Fiscal Year 2022)
|Budget Amount *help
¥2,860,000 (Direct Cost: ¥2,200,000、Indirect Cost: ¥660,000)
Fiscal Year 2023: ¥1,430,000 (Direct Cost: ¥1,100,000、Indirect Cost: ¥330,000)
Fiscal Year 2022: ¥1,430,000 (Direct Cost: ¥1,100,000、Indirect Cost: ¥330,000)
|voice privacy / time scale modification / phase vocoder / speaker anonymization / gender / HCI / speech security
|Outline of Research at the Start
This study mainly aims to develop a privacy-aware computing system for assisting speech communication. Unlike most existing systems that only focus on performance accuracy, this study addresses the protection of voice privacy in system development by a novel speaker anonymization method.
|Outline of Annual Research Achievements
In this fiscal year, we proposed speaker anonymization methods based on time scale modification (TSM) algorithms. The study finds that using the phase vocoder-based TSM method is more suitable for speaker anonymization due to the human voice's harmonic structures. The proposed method balances privacy and utility metrics better than baseline systems. Besides, we also analyzed the effect of anonymization on the perception of gender by utilizing a gender classifier model that was built using x-vector speaker embedding. The results of our study were presented at the Voice Privacy Challenge 2022, joint with the Interspeech 2022 conference and the 14th annual conference organized by Asia-Pacific Signal and Information Processing Association 2022.
|Current Status of Research Progress
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
The progress of this project is going well as planned. The speech analysis has been performed to obtain the features related to personally identified information (PII). We investigate pitch shifting using two major categories of TSM algorithms for speaker anonymization. Our recent finding from this study is that the human voice contains harmonic structures; thus, applying PV-TSM, which is more suited to a harmonic component, could benefit speaker anonymization. Subsequently, the phase adaptation may manipulate not only fundamental frequency but also the PII-related acoustics features. Our method outperformed the x-vector-based speaker method, which has limitations in its complex training process, low privacy in an a-a scenario, and low voice distinctiveness.
|Strategy for Future Research Activity
In the currently proposed methods, several remaining issues exist. For instance, the speaker anonymization target needs to be clearly defined. As a result, the application for speaker anonymization has several limitations on the attack models. In the future, the development of more secure and robust speaker anonymization with attack models will be the focus. Hence, it can be applied for broader applications. Important ethical and privacy concerns will also be considered when developing speaker anonymization techniques.
Report (1 results)
Research Products (6 results)