I’ve been having a play with the free trial of Kaleidoscope software from Wildlife Acoustics. This was partly prompted by the courses I’ve been running on behalf of BCT for some of the larger consultancies who generate far more data than they can ever analyse manually, so they are very keen to embrace automated processing software. To test it I’ve run through some 10,000 files that I’ve manually looked through so I know what’s on them. These are mostly from a Pettersson D500x, but there are also a large number of files from the SM2BAT+.

The basic interface is modular, in that the file conversion utility is free. You can use this to batch process and convert .wac files to .wav and also zero crossing files. To this you can add the Kaleidoscope viewer (£325) or opt for the Kaleidoscope Pro version which includes the viewer and a set of classifiers for UK bats (£755). You can download and try Kaleidoscope pro for 15 days to try it out, which is what I did.

To start with, there has clearly been a lot of time spent on the interface. It works nicely, is simple and does the job. One minor irritation is that you can select an output directory that is above the data directory. In other words, if you data is in C:\field_data, you can’t put the results either in that directory, or in one above it such as C:\field_data\results. They have to go into a directory at the same level or below the one you are processing. If you are just processing files to classify them, don’t check any of the output file formats unless you want copies of the processed files in your results folder. With 60 GB of data to process, this can cause havoc!

If you are classifying calls, you have the option of switching in and out any of the current 11 species classifiers. Myotis is still lumped together (sensibly), and the newest addition is a classifier for Nathusius’ pipistrelle. There are two classifiers for lesser horseshoe, a normal one and one for the 192 kHz recording mode where lesser horseshoe calls would be aliased.

To start, you select the directory on the left hand side of calls you want to process.  You can also include sub-directories from this which is a nice feature meaning you can batch process A LOT of files. You select or make an output folder on the right hand side for the output files, select the classifiers you want to use (i.e. the species you want to look for), hit ‘process files’ and go away for a cup of tea, lunch or a mini-break to a top European destination depending on how many files you need to process. For my files it took around 10-15 minutes per night of data, which for this type of processing is very very fast. However, as this is very CPU intensive, I couldn’t use my PC for anything else while the processing was going on. If I did try, Kaleidoscope threw up a few file I/O errors claiming that the file was too big to process. So best to leave it to do its job. If you have a lot of detectors for a lot of nights, this could take you a few days, so be mindful of that.

The output is presented as a table, but also output as a .csv file to your selected results directory. It tells you the file name, species ID, number of pulses and certainty of identification. You can then click on each entry and that file opens up in the viewer which is a really nice feature so you can check it and if necessary amend the identification. You can also click on the column heading to sort by species to look for anything that is not a pipistrelle. I have to say I’m not a great fan of the layout and look of the viewer, but then I’ve used Batsound for years so am more used to that. I’m sure you can customise the viewer to suit the look and layout you want.

So how did it fare on identification? Well, I have to say it did pretty well considering. It’s not overly keen to classify things, which I like. I have used other software which is intent on suggesting the most bizarre species. Nothing too odd came out of this, and it did seem to deal quite well with the dreaded pipistrelle social calls which other software turns into barbastelles among other things. There was a category of ‘noID’ which is where the software admits it can’t identify something, which again is pretty honest.

It is quite keen to pull out P. nathusii (I downloaded it with the nathusii classifier). So a P. pipistrellus with FmaxE at 41 kHz was a nathusii, with high confidence which it clearly was not. This happened quite a lot where groups of common pipistrelles were flying together and shifting their call frequencies away from one another. Personally I wouldn’t take any nathusii classification at face value without checking it manually.

It struggled with the big bats group. Obvious N. noctula were right, but some serotine/leislers FM-QCF calls were attributed to N. noctula when they were too high for my liking. Also some higher QCF noctule calls at 22-23 kHz were attributed to N. leisleri when the IPI was far too long to be leisleri. Again, I’d like to have a look manually at all the big bat classifications.

One bizarre occurrence was that high amplitude recordings of P. pygmaeus which had nice harmonics at 110 kHz were picked up with 70% plus confidence as lesser horseshoes rather than pygmaeus, even though they were lovely pygmaeus recordings.  P. auritus also cropped up a few times in the recordings but it was usually a Myotis, so that needs watching. It did a pretty decent job of ignoring pip social calls rather than attributing them to something, again which is pretty good.

The software did fail to deal with recordings which had more than one species in them, but that’s not terribly surprising.

One potentially more alarming issue was that I had about five nights of data from inside a church which had a brown long-eared roost in it. The software failed to find a single long-eared call even though there were plenty of classic (weak) calls in the recordings, as well as the usual social calls. This happened even when I turned the sensitivity up on the classifier (there is a more accurate/more sensitive switch).  I wouldn’t be happy using this to find long-eared bats in barn conversions for example without manually going through the data. I even edited out good long-eared recordings and put individual files through. It wasn’t that it didn’t classify them, it just couldn’t find them.

In summary, it did a pretty good job as a pipistrelle filter, and can be great for sorting out other species to check manually, but I’d still want to have a look at any file that wasn’t a P. pipistrellus/pygmaeus to check the classification. The software can save a great deal of time, and in some larger scale surveys may be a necessity, but I would be reluctant to accept any non-common or soprano pipistrelle identification at face value without having a look at it first.