Startle Detection

Code

• msdetect.m (MATLAB function)

Example

Steps of Algorithm

Based on the startle detection prorgram by Jennifer Healey (Dec 13, 1999)

1. Filter out high-frequency noise
   

   (Optional)

2. Find significant startle regions
   

   A startle region is first roughly defined as a "large" region where the first sample point has the only "significant" positive first difference, and no other point in the region has such a "significant" positive first difference.

   • "large" region means >31 sample points by default
   • "significant" first difference means >0.003 by default

3. Fine Tuning
   

   Given the startle regions, we define

   • the beginning of a startle (beg) to be the first sample point before a region with a negative first difference, and
   • the end of the startle (end) to be the maximum point between the beginning of this region and the beginning of the next.
   • The duration of the startle is then end-beg, and
   • the magnitude of the startle is s(end)-s(beg), where s is the signal.