May 17th 2013 (Fri)
NEW DATE: HRC Seminar with Oded Ghitza

Dr. Oded Ghitza, Research Professor, Hearing Research Center and Department of Biomedical Engineering, Boston University


Title: Speech perception, cortical theta oscillations and auditory channel capacity

Abstract:

Speech, time-compressed by a factor greater than about three is unintelligible. Surprisingly, intelligibility is considerably restored by 'repackaging' – a process of dividing the compressed speech signal into segments, here called packets, and inserting gaps between the packets (Ghitza and Greenberg, 2009).  Periodic repackaging is defined by two parameters, the packet duration and the packaging rate. The amount of information, in bits, carried by the packet can be inferred from the packet duration, and the information rate, in bits/sec, is determined by a mix of packet duration and packing rate. 

This talk is concern with the following questions: (i) what is the maximum speech-information rate to be perceived without error (i.e., what is the auditory channel capacity, in bits/sec)? and (ii) what is the cortical function that determines auditory capacity? To address these questions, we measured intelligibility of naturally spoken 7-digit telephone numbers time-compressed by factors up to eight, with parametrically varied repackaging parameters. 

The results show that, at capacity, packaging rate and packet duration are in correspondence with properties of cortical theta oscillations. For any prescribed compression factor, packaging rate is 9 packets/sec – in correspondence with the upper limit of cortical theta (9 Hz). The information delivered by the packet is the information contained in one uncompressed intervocalic segment (one theta-syllable; Ghitza, 2013). These estimates are at capacity because, for any other packaging-rate/packet-duration combination with higher information rate, intelligibility deteriorates.


Seminar Information

Seminars are generally held on Fridays at 10:30 AM in Room 203 of 44 Cummington Street. (Exceptions in time or location will be noted for any talks that are different).

Close