You are here

Audio Calibration Section - Under Construction

Error message

  • Deprecated function: implode(): Passing glue string after array is deprecated. Swap the parameters in drupal_get_feeds() (line 394 of /homepages/14/d296057714/htdocs/drupal/includes/common.inc).
  • Deprecated function: The each() function is deprecated. This message will be suppressed on further calls in menu_set_active_trail() (line 2396 of /homepages/14/d296057714/htdocs/drupal/includes/menu.inc).

The solution is to analyze the audio signals in a way analogous to the human ear and brain. My argument is that most of the elements of psycho-acoustic models used by lossy audio compression can be duplicated by modeling the physical processes occurring in the human ear. By modeling the action of the stereocilia (hair cells) within the inner ear as damped harmonic oscillators many of the features of psycho-acoustic models are reproduced. For example, the frequency dependence of time masking is a natural consequence of the decay rates of the oscillators.

The signals analyzed in this way are then correlated in order to compare the similarity. The resulting correlation vs. time is then statistically analyzed for similarity to public listening tests. The original development used Roberto Amorim's listening tests, but they are no longer available. The final refinements relied heavily on one of the more recent tests by Sebastian at HydrogenAudio.

I found that P99 (the correlation value which is the upper limit of 99% of the values) correlates the best with human listeners. In the Sebastian test the correlation was usually over 80%. However, a few examples showed listeners were able to discern a difference when the P99 values showed no significant difference. I found that in those cases that the correlation value was fluctuating erratically on the sample judged to be inferior. The best statistic that captured this feature was the residual median smoothed root mean square value (median smooth the data and then calculate the RMS of the difference between the raw and smoothed data).

So the results of Sebastian's tests could be recovered by using the P99 and RMS Residual Median. A significant difference in either statistic is judged to be significant.