BME MS Thesis Defense - Miguel D. Cepeda

11:00 am on Thursday, April 11, 2013
1:00 pm on Thursday, April 11, 2013
677 Beacon Street, room B02 (CompNet Auditorium)
Committee Members:
Barbara G. Shinn-Cunningham, PhD, Professor of Biomedical Engineering (Advisor, Chair)
H. Steven Colburn, PhD, Professor of Biomedical Engineering
Gerald D. Kidd Jr., PhD, Professor of Speech, Language and Hearing Sciences

Title: "Effects of spatial separation on across-frequency grouping in narrowband speech"

Understanding how we perceive speech in the face of competing sound sources coming from a variety of directions is an important goal in psychoacoustics. In everyday situations, noisy interference can obscure the content of a conversation and require listeners to integrate speech information across different frequency regions.
In this thesis, two studies investigate the effects of spatial separation on the grouping of two spectrally separated, narrow bands of target speech with a variety of filler stimuli centered in between these bands. Target sentences taken from the IEEE corpus were broken into two -octave bands with the lowest centered around 370 Hz and the highest centered around 6 kHz. The first study explored the spatial influences of spectral restoration, the enhancement of intelligibility in spectrally degraded speech induced by noise. The primary experiment measured speech intelligibility of the speech bands (presented diotically) with a single band of noise between 700 Hz and 3 kHz used as the filler and then with the same noise band modulated by the target speech envelope as the filler. These fillers were presented diotically as well as with an ITD of 600 s leading to the left ear. Performance was worse for the unmodulated, spatially separated noise condition than the other conditions with noise. The second study explored the effect of attention on intelligibility of speech bands presented from the left with related fillers. The fillers used in this study were dual bands of vocoded or narrowband speech presented either from left or right. The fillers were derived from either the same target speech token (matched) or an independent sentence (conflicting). In a key experimental block, listeners were instructed to attend to the target speech on the left while either conflicting bands or, infrequently, matched bands were presented on the right. The infrequently presented matching trials were physically identical to trials in another block where listeners were instructed to attend to both ears. Results showed that splitting the target and filler across the ears degraded intelligibility, however, directed spatial attention had no effect on performance. These results demonstrate that speech elements group together strongly, overcoming spatial attention, even for degraded speech.