ICSV26

One of the most important things we do every day is understand spoken language. Identifying spoken words requires the rapid integration of multiple acoustic-phonetic cues in the richly multi-dimensional and transient speech signal. Every meaningful difference in sounds (e.g. the difference between the sounds at the beginning of 'peach' and 'beach') is signalled by multiple acoustic-phonetic cues (e.g. spectral and temporal differences) and often a single cue is not enough for unambiguous perception. Furthermore, the speech signal is notoriously variable and this variability stems from many sources including the talker (e.g. differences in vocal tract size, differences in dialect) and the context (e.g. the immediately following sound, speaking style). This variability and multidimensionality in the signal are part of the central challenge that human speech perception must solve and compared to automatic systems, humans are better able to deal with the variability in particular. This talk will present an overview of challenges related to multidimensionality and variability in spoken word recognition, including sources of variability, and what is known about the solutions, including the role of plasticity. It will also introduce some recent work in my lab investigating differences between individual talkers and listeners.