https://github.com/macaodha/batdetect Well I've got it running on with Anaconda 3 and it scanned 480 files and flagged 40 odd as containing bats calls - which is looking correct so far. What I'm failing on is getting it to look for the wave files on an external drive ie f:\, anyone know if you can only have the files as a subdirectory or am I missing something simple? should I have the python files on the F: drive?
top of page
To test this feature, visit your live site.
6 Comments
bottom of page
Also Batclassify will run when batdetect falls over complaining about "corrupt" files - I may have lets the batteries run too low <shamed face>
I've been trying batdetect and also BatClassify. I have some success with Batclassify, running at 384 kHz, 15 second file length. It does assign Ppyg social calls to NSL ( and has no Pnat) model. Batdetect runs faster but I haven't had much of a chance to get into the data. If I find the time I'll compare the two. I did find that BatDetect (anecdata alert) was not great at picking out all the calls from a file and was very prone to false positives. For what I am doing at the moment I take the spreadsheet of the 2500 or so file results from batclassify and rank them by score for each species in turn. I can then manually examine files down to a cutoff value (species dependent) to see if they are correct. Typically anything over .8 is definite for pips but you are not sure if there is something else in there, and I had some very faint echolocation with loud social calls that flagged up as NSL. For a moment I thought there was a Leislers in Scotland but more careful examination showed it to be a ppyg social call.
Batdetect takes an hour or so to run on 2500 files on my not very high spec laptop and requires some python skills. Batclassify takes about 6 hours (ie overnight) for each directory of about that size.
Thanks, unbelievable timely as this morning's job is to scan some cards...
I've been trying this. I can happily set a mounted drive by setting it to "E:\\Path/to/my/data/"
and it then runs fine. I have fettled it to run under Python3.
The quality of the output is variable. The audiomoths have periodic noise that is interpreted as a lot of calls. Depending on the species you are looking at, you may want to post-process the output and reject assigned calls that are too close together. My thinking is that a new set of models need to be developed for the AudioMoth, or the data needs some pre-cleaning to denoise it.
I should have some time to return to this after the weekend. I now have some recordings in a much quieter place so will try to run those. Depending on how you are running your script, you may find that the drives are mounted under /f/ instead of F:.
That's interesting. I got an horrendous error rate with the files I was using but perhaps they had more noise than expected. (under a road bridge over a river in the rain). I've been discussing it with the authors.