Aculab to study inner workings of automatic speaker recognition with University of York
Aculab has announced a collaboration with the University of York on a language and linguistic science research project titled Towards linguistically-informed automatic speaker recognition, to examine how speaker-characterizing information captured by automatic speaker recognition (ASR) systems maps to linguistic properties of the voice. The study could have implications for the development and improvement of ASR, enhance public and judicial understanding of ASR, and improve voice analysis for forensics.
The project is funded by doctoral training partnerships program WRoCAH (White Rose College of the Arts & Humanities). It will address the extent to which ASR systems capture a voice’s tangible linguistic properties, whether understanding the information captured by ASR systems can enable predictions about what speakers will be problematic for them, and whether linguistic information can be used to improve ASR performance.
Despite the low error rates of advanced ASR systems in experiments, little is known about what information they actually capture, Aculab says. Meanwhile, the technology is increasingly used in forensic applications and legal settings.
“We will be delighted to host the successful student and to provide unique employment-related skills via experience in a commercial environment,” explained Dr Ladan Baghai-Ravary, Speech Technology Director at Aculab. “This collaborative project will allow us to address new forensic applications of our technology, and in general it will increase the uptake of automatic speech and voice analysis within the forensics industry, by providing a legally acceptable interpretation and justification of the evidence that such systems can give.”
Understanding and modelling the human voice is important for humanities such as linguistics and phonetics and sciences such as engineering and computer science, according to the announcement. Different disciplines approach the issue in different ways, and little has been done to explore the intersection between disciplines, however.
“The collaboration between York and Aculab is unique,” stated Dr Vincent Hughes, Department of Language and Linguistic Science, at the University of York, “and we are very pleased to work with a leading commercial developer of ASR systems.”
Hughes says York hosts one of the world’s leading research centers for forensic speech science, and the project is one of the only large-scale investigations of its kind.
Aculab Product Manager Ian Colville recently shared insights on how to secure solutions with voice biometrics in a Biometric Update guest post.