Acoustic, perceptual, and cognitive features for objective speech quality and intelligibility prediction: Dr. Tiago Falk

Acoustic, perceptual, and cognitive features for objective speech quality and intelligibility prediction: Dr. Tiago Falk (INRS - EMT). A CRBLM symposium

Abstract:

Subjective listening tests are commonly used by telecommunications service providers to gauge network performance; by speech language pathologists to characterize speech disorder severity or to document treatment outcomes; or even by clinicians looking to detect diagnostic “markers” in e.g., autism spectrum disorders (ASD). Subjective listening tests, however, are costly, laborious, and subject to many intrinsic variables and biases due to e.g., familiarity with the patients and their speech. Objective quality prediction, on the other hand, replaces the human listener panel by a computer algorithm, thus provides an economical and reliable (repeatable) alternative.

This talk will describe our efforts at developing biologically-inspired signal processing algorithms that extract relevant acoustic and perceptual features for automated speech quality and intelligibility assessment. Three applications of the developed algorithms will be presented. First, I will present findings on an auditory-inspired quality model for natural speech which was recently shown to be particularly useful for cochlear implant users in reverberant environments. Second, I will show that by characterizing and integrating multiple dimensions (e.g., prosody, temporal dynamics, vocal source excitation), it is possible to predict the intelligibility level of a given dysarthric speaker as well as predict the success of such speaker in using an off-the-shelf speech recognizer for communication purposes. Lastly, I will present some pilot findings on the use of automated acoustic feature analysis for the purpose of diagnosing ASD risk in non-verbal toddlers. Time permitting, I will describe an ongoing project which aims at developing“cognitive” features for objective speech quality perception, motivated by some recent findings in our brain-computer interface research.

About Dr Tiago Falk:
Tiago H. Falk received the BSc degree from the Federal University of Pernambuco, Brazil, in 2002, and the MSc and PhD degrees from Queen's University, Canada, in 2005 and 2008, respectively, all in electrical engineering. In 2007, he was a visiting Research Fellow at the Sound and Image Processing Lab, Royal Institute of Technology (KTH), Sweden, and in 2008 at the Quality and Usability Lab, Deutsche Telekom, Germany. From 2009-2010 he was a Postdoctoral Fellow at Holland-Bloorview Kids Rehabilitation Hospital, affiliated with the University of Toronto. Since Dec. 2010, he has been an Assistant Professor at the Institut National de la Recherche Scientifique (INRS‐EMT) in Montreal where he is Director of the Multimedia/Multimodal Signal Analysis and Enhancement (MuSAE) Laboratory. Dr. Falk is also an Adjunct Scientist at Holland-Bloorview and a Research Advisor for InteraXon Thought Controlled Computing. His research interests include speech quality measurement and enhancement, neural correlates of speech quality perception; assistive technologies; speech communication disorders; and human‐machine interfaces. He has published over 70 journal manuscripts, conference papers, and book chapters in these topics.

More about Dr. Falk: http://www.inrs.ca/english/tiago-h-falk