The BTO Acoustic Pipeline is free to use up to 100 GB / year (about 50 nights of triggered bat recording). I am not un-biased to comment on performance because I built the classifiers that are used by the Pipeline, but I would strongly recommend that you take a dataset (with everything - not just pulling out the best bat recordings) that you have taken time to manually check, put it through each option that you are considering, and critically compare the performance. 

In particular, the largest differences in performance between a good and more poorly performance classifier will be for some of the more cryptic species, so particularly focus on e.g. Myotis species, Plecotus species - if you are in an area where Brown and Grey long-eared bat are present. See how well the approach is able to identify multiple species if present in a wav file - is the classifier just identifying the species with the strongest calls, how well does it identify bat social calls / e.g.  is it mis-identifying Pipistrellus social calls as Nyctalus species , how well does it identify weak bat calls in noisy recordings (e.g. does it miss Barbastelle calls in recordings with bush-crickets), does it mis-identify small mammals calls as bats e.g. Brown rat calls as Nyctalus species. 

Another option for you could be Arbimon - it's free and does not require any coding/programming knowledge! 

Hey Kate I don't have any explicit ideas at the moment but will think on it - such a cool concept! I wonder if there's some form of tactile spectrogram that you could develop, or a way that you could have them try to draw a representation (so assign some shape) of what they are hearing as a means of classification a call and then clustering those shapes? There's a way of creating an image using foam and paper I believe so they could have a tactile record of those shapes. Guess it depends what you want to the quantitative results to be. Super cool that you are looking for greater accessibility in acoustics!!!

Hi Kate, 

We've developed a screen-reader workflow for a classification problem on our "Ocean Voices" Zooniverse project, which simply asks folks to label sounds based on what they hear and omits the spectrogram altogether. There are lots of screen-reader users who are active in the Zooniverse Talk forums, so they may have valuable input for you as well. 

Once a person has labeled data, I wonder if they could run automated detectors over the data in Pamguard, calculate features (using something like the R package PAMpal), and then use the BrailleR package to explore the statistics in R. This article has a pretty interesting summary about statistical software for visually impaired folks - might not be news to your student, but I thought it was pretty cool. 

I'm very curious what our friends who are visually impaired might notice in the acoustic data. Best of luck to you and your student! 

Cheers, 

Anne 

Hi Kate- ARISTA Lab (Advanced Research in Inclusion & STEAM Accessibility) is actively working on this through their eclispse soundscape project. I recommend reaching out to MaryKay to get the latest on their project.  

Megan