Apurahat ja residenssipaikat

Tiede

Identifying stable and changing acoustic features in intentional voice change in Finnish speech

Hakemuksen tiivistelmä

It remains a significant challenge to automatically recognise speakers who intentionally alter their voices to conceal their identities, despite the substantial progress made in automatic speaker verification systems. The intentional alteration of voice is often employed in situations such as blackmailing calls or robberies, where the offender wants to remain anonymous. One crucial step toward achieving robust automatic speaker verification systems is gaining a comprehensive understanding of how speakers modify their articulatory configurations (and subsequently, the resulting acoustic characteristics) when they change their voice. In this project, I aim to explore the speech features that are systematically altered or remain stable during intentional voice changes, specifically when speakers attempt to sound like either younger or older individuals. My focus will be on intentional voice change features in Finnish speech, utilizing a publicly available corpus that includes modal and disguised speech with age modification. The outcomes of this project are expected to have significant implications for developing reliable automatic speech verification systems, enhancing the security of voice-activated systems, forensic applications and for basic research.

I was awarded a one-year postdoctoral grant and a travel grant from the Kone Foundation to conduct a research project (1 October 2024 – 30 September 2025) investigating how Finnish speakers intentionally modify their voices to sound younger or older, with the goal of understanding the acoustic changes involved in deliberate voice alteration.
Using a Finnish speech corpus containing both modal and intentionally modified voices, I examined which speech features—both at the segmental level and in global acoustic measures—change during intentional voice modification. The results revealed systematic patterns in certain features, such as vowel space dispersion, when speakers attempted to sound like a child, whereas changes were more variable when imitating old voice. Analyses using a state-of-the-art automatic speaker recognition system showed that modern systems still struggle to identify speakers who deliberately alter their voices. These findings highlight acoustic factors that contribute to recognition failures and may guide future improvements in speaker verification systems. They also provide new insights into intentional speech changes, with implications for forensic phonetics and speech science more broadly.
Building on these results, I aim to extend this study using electromagnetic articulography to investigate the articulatory strategies underlying intentional age-related voice modification, providing a deeper understanding of the actual physical articulatory mechanisms as well as the resulting acoustic changes in speech.

Takaisin apurahalistaukseen