Description |
Committee members:
Prof. H. Steven Colburn, PhD (Advisor, Chair)
Prof. Gerald D. Kidd Jr., PhD (SLHS)
Prof. Herbert F. Voigt, PhD (BME)
Prof. Kamal Sen, PhD (BME)
Title: “ A study of detection models for narrowband reproducible noise”
Abstract:
Binaural information is essential for better performance in tasks such as redirecting attention toward spatially distinct acoustic sources in reverberant environments. Binaural hearing studies have focused on how the binaural system can improve the extraction of information from one source in the presence of competing sources. Devising models that reveal performance trends analogous to participants’ behavior observed in experiments is an efficient way to test our understanding of binaural hearing. The ability to detect a pure tone in the presence of a noise masker is known to depend on the configuration of the noise waveform. Repeated presentations from a controlled set of masking waveforms provide the capability to precisely characterize detection performance at the individual noise sample level.
This thesis follows up on the reproducible noise masking experiments conducted by Scott K. Isabelle’s in his graduate thesis work. In Isabelle’s work, probabilities of detection (hit rates or Pd) and false alarm (Pf) were estimated for each of the 30 noise waveforms. The noise waveforms were generated from a one-third-octave Gaussian noise process centered at 500 Hz. The presented noise was identical to both ears. The target in the detection experiment was a 500-Hz pure tone, presented interaurally out-of phase. All measurements were in the N0Spi condition. In Isabelle’s study, a number of binaural models were tested by measuring the correlation between hit rates and decision variables for each waveform according to the assumptions of each model. Some models such as those based on variability in interaural time or intensity differences were statistically supported. However, the variation of false alarm rates across noise samples was not explained by these models based on interaural differences. This is because the variations of the decision variables from these models are all zero for the noise-alone condition. |