Abstract Swedish |
The time-frequency (TF) spectral representation of Auditory Brainstem Response (ABR) signal data<br>
provides information about their spectral contents. We apply the Spectrogram, Thomson Multitaper and<br>
Peak Matched Multiple Window (PM MW) spectral estimation methods to four different number of clicks<br>
per average (i.e., 1313, 300, 100 and 50 number of clicks per average) of a simulated signal data. For the<br>
purpose of model selection we simulate sinusoidal signal data which have the same trend as the empirical<br>
ABR signal data, and then apply the selected model to ABR data from 17 healthy, normal hearing individual<br>
ears as recorded using SD-BERA, SensoDetect-Brainstem Evoked Response Audiometry. The root mean<br>
square error (RMSE) is the main tool used to compare the proposed spectral estimation methods. The<br>
Spectrogram is found to be an appropriate method of spectral estimation for signals with relatively low<br>
disturbance. In particular, for signals with a white disturbance with standard deviation, , value in the<br>
interval 0,15.0, it is found to be best of the three methods. For 15.0 ≤ ≤ 30.0, the PM MW method<br>
performs as good as the spectrogram, if not better. Finally, for ≥ 30.0 the PM MW continues to be the<br>
best of the three methods where as the Spectrogram turns out to be worst of them. |