Workshop on Real-world speech perception: challenges, consequences and models


10.00 to 13.30, 25th October 2017

Salón de Actos (ground floor)
Micaela Portilla Research Centre
UPV-EHU, Vitoria-Gasteiz


Chair: Prof. Maria Luisa Garcia Lecumberri


Effects of cognitive load on speech perception

Prof. Sven Mattys, University of York, UK

Improving the validity of speech-recognition models requires an understanding of how speech is processed in everyday life. Unlike listening conditions leading to a degradation of the signal (e.g., noise), adverse conditions that do not alter the integrity of the signal (e.g., cognitive load, CL) have been under-studied. Drawing upon behavioural and imaging methods, our research shows that CL reduces sensitivity to phonetic detail and increases reliance on lexical knowledge. The results not only constrain our understating of the functional architecture of speech-recognition models, they also invite a re-analysis of the validity of standard hearing tests for assessing everyday listening.


Speech intelligibility prediction

Dr Cassia Valentini-Botinhao, University of Edinburgh, UK

In this talk I will present a few different objective measures of intelligibility, describe how they differ from each other and how they perform under different conditions. I will present some of my work using one those measures to improve intelligibility of synthetic and natural speech in noise and discuss their limitations in that context.

Download the slides for this talk


Measures and models of microscopic speech perception

Dr Johannes Zaar, DTU Copenhagen, Denmark

Speech intelligibility in a given acoustic condition and/or listener group is typically determined by presenting sentences to listeners and counting the amount of correctly recognized speech items, which provides a global measure of speech perception. To investigate the relation between the fundamental speech cues and the resulting speech percept in more detail, speech perception can also be assessed microscopically by presenting nonsense syllables (e.g., /ta, ba/) to listeners and evaluating the responses in terms of phoneme recognition and phoneme confusions. In this talk, several studies focusing on factors that influence consonant perception in terms of the source (stimuli), the transmission channel (noise, hearing-instrument processing), and the receiver (listeners) will be presented. Furthermore, a stimulus-driven microscopic speech perception model will be introduced to predict the perceptual data. The utility and limitations of measuring and modeling speech perception at the microscopic level will be discussed.


This event is supported by the EU Marie Curie European Training Network “ENRICH”. Attendees from outside the network are welcome, but please contact the ENRICH Project Manager Mónica Loyo ( to confirm attendance.