THESIS
2015
xvi, 115 pages : illustrations (some color) ; 30 cm
Abstract
In a recent industrial consultation study, the author observed that workers could easily detect a
59 dBA train alarm in the presence of 77 dBA train noise; a signal-to-noise ratio (SNR) of -18
dB. A literature review indicated that little is known about alarm detection with SNRs of -18 dB
or below. The first experiment of this study was designed to repeat the observations with 12
participants. Four stimuli conditions with alarms at -18, -21, -24 and -∞ dB SNRs were studied.
The train alarm (2 kHz with harmonics at 4 kHz, 6 kHz, 8 kHz and 12 kHz) and the train noise
(with pink spectrum) were similar to the observed signals. Results indicated that the subjects
were able to detect the train alarm with -24 dB SNR and the perceived loudness of the train
alarm was significantly diffe...[
Read more ]
In a recent industrial consultation study, the author observed that workers could easily detect a
59 dBA train alarm in the presence of 77 dBA train noise; a signal-to-noise ratio (SNR) of -18
dB. A literature review indicated that little is known about alarm detection with SNRs of -18 dB
or below. The first experiment of this study was designed to repeat the observations with 12
participants. Four stimuli conditions with alarms at -18, -21, -24 and -∞ dB SNRs were studied.
The train alarm (2 kHz with harmonics at 4 kHz, 6 kHz, 8 kHz and 12 kHz) and the train noise
(with pink spectrum) were similar to the observed signals. Results indicated that the subjects
were able to detect the train alarm with -24 dB SNR and the perceived loudness of the train
alarm was significantly different among all four conditions. The SNR level of -24 dB was further
reduced to -28 dB (median value) in the second experiment with sixteen participants using a two
interval forced choice (2IFC) testing protocol. This was referred to as the alarm-in-noise (AIN)
detection threshold. In addition, results indicated that if free-field spatial information was
removed by presenting the train alarm and the train noise monaurally and diotically, the median
AIN levels raised back to -13 dB and -14.3 dB, respectively. There was a statistically significant
difference between the detection thresholds collected at the monaural and free-field conditions
(p<0.05). In the second experiment, train alarms of different durations were tested and found to
have no significant effect on the detection thresholds. A large inter-subject variability was also
observed ranging from -9 dB to -46 dB SNR. In an attempt to uncover the mechanism behind an
AIN detection threshold of -28 dB SNR (or -46dB in one listener), a customized biologically
inspired model for hearing was constructed. The model was based on the existing Matlab
Auditory Periphery (MAP) model developed by Meddis (2006). It contains basic modules to
simulate and predict sound transmission from the pinna to the middle ear, the cochlea and the
subsequent excitations of auditory nerves. The simulation results of the initial model indicated subtle but repeatable changes in the basilar membrane (BM) displacements and auditory nerve
firing patterns related to the stimuli conditions of the experiments. In particular, the effects of
efferent feedback of the medial olivocochlear system (MOCS) were simulated because this type
of efferent feedback had been shown to improve speech intelligibility in noise. A third
experiment was conducted to test the hypothesis that the MOCS efferent feedback would help
the detection of train alarm in loud train noise. Subjects of the second experiment were tested for
their strength of efferent feedback in terms of their contralateral suppression of the transient-evoked
otoacoustic emissions (TEOAEs). To our surprise, correlation analysis indicated a
significant (p<0.05) but negative correlation between contralateral suppression of TEOAEs and
masked detection thresholds. This implied that stronger MOCS efferent feedback worsens the
detection performance in negative SNR conditions. The fourth experiment verified and
confirmed that the negative SNR levels of AIN also held for tone-in-noise (TIN) detection
thresholds (1 kHz and 2 kHz). The thesis contains novel and original results that are beneficial
for both industrial and academic purposes. The customized and calibrated biologically-inspired
MAP models can be a useful platform for testing and developing future auditory alarms in noisy
environments.
Post a Comment