Recognizing train passage

[Bill Gearhiser]

Hi, folks. I wonder if I could pass my project by you to see whether I've selected the wrong software.

I need to be able to identify and time-stamp the passage of trains about a half mile away. Basically I would run the software, listening to an outside mic, 24 hours a day. It's not necessary to identify the individual trains; I just need to know when they passed. This is made easier by the fact that they use the familiar long-long-short-long horn pattern at a nearby intersection. It's complicated by the fact that the speed of the pattern varies with the train's operators, the train horns are all different pitches; and doppler shift plays games with the horn pitches.

Do you think echoprint is the right way to go, or is it too strict -- requiring me, for example, to have recording of all the different locomotives to compare against?

Thanks.
Bill in Florida

[Dan Ellis]

Bill -

Fingerprinting relies on very precise matches in frequency and timing,
so won't work with the small differences between train horns that you
describe. But you might be able to detect the horns with something
like a matched filter looking for the specific harmonics of the horn
sounds - take an FFT over a fairly long window, like half a second,
then look for sharp peaks in the magnitude around the pitches of the
horn. Maybe make some test recordings and look at the spectrograms in
Audacity or similar to see if you can visually detect anything.

DAn.

> --
> You received this message because you are subscribed to the Google Groups
> "echoprint" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to echoprint+...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.