search button
For hearing aid users    |    Contact    |    Publications
 
proproductsfittingsystemsinnovationspaediatricsaboutwidexwidexlearn

Percentiles

A central element in the statistical analyses performed in Widex hearing aids is percentile levels.

Percentile levels provide information on the level distribution, that is, how the loudness level of the incoming signal changes over time. When obtained for multiple frequencies, this information provides quite a detailed picture of the sound environment.

The figure below illustrates how the level of the sound changes over time. In the beginning it is soft, then it becomes a little louder, then soft again, then very loud, and finally soft again.

Widex illustration
Illustration: Changes in the loudness level of the signal over time

The level histogram below shows how often each level occurred in the signal over time. The sound level has been measured at regular time intervals, and each loudness level’s frequency of occurrence has been plotted.

Widex illustration
Illustration: The x-axis shows the loudness level (the leftmost column shows the softest level). The y-axis indicates how often each level has been measured.

One piece of information that would be relevant to extract is precisely how soft the soft parts of the signal are and precisely how loud the loud parts are. This is where percentile levels can be helpful.

A given percentile level indicates how much of the signal was below or equal to that particular loudness level. For example, the 10% percentile level is the level at which the signal was louder 90% of the time.

Thus, if we identify the 10 most quiet percent of the measured level, we have a good indication of how quiet the quietest part of the signal was.

Widex illustration
Illustration: The 10% percentile level indicates how quiet the quietest part of the signal was


Similarly, if we want information about how loud the signal becomes, we could extract 90% of the softest measurements. In other words, we could extract the 90% percentile level where the signal is louder only 10% of the time. A high level percentile such as the 90% percentile will provide information about the peak level in the signal.

Widex illustration
Illustration: The 90% percentile level indicates how loud the signal was


In this way, percentile level estimation can provide statistical information about the ambient sound environment which can be used at different stages in the processing sequence to make sure that the right amount of gain is provided for the specific user in the specific situation.