Can anyone recommend a usable resource/tool for species detection from acoustic data please? A middle ground between a phone app and something like Arbimon would be about the level. Briefly, a phone app lacks flexibility in data collection, i.e. it cant be left out all night or left running for long periods. However, Arbimon is not that useful to anyone below ecologist level, as it only tells the user what species is present if the user completes their own validation, i.e. the user has to identify all species themselves. I'm looking for something that can analyse data from an AudioMoth, uploaded by citizen scientist participants, and actually identify species.
There are specialized centres in USA and UK they do this job. I suggest to you send your specimen to one of the centres, and you probably you find other centres in other countries such as Germany, China or India. Wish you luck.
Jamie, sounds like an interesting project. As a forethought, I know the term "detection" is commonly used in this context, but it may be more useful to think of the problem in terms of recognition. The reason I say this is that a lot of work has been done in developing speech recognition systems, and you may find this work helpful in formulating an approach to your problem. Your problem is potentially far more complex however than speech recognition, as speech recognition has to deal only with 50 or so phonemes (the building blocks of words) while the number of species and their identifying sounds is likely to be much larger. To make this problem tractable, you will probably need to limit the number of species you are trying to recognise, either by geographic limitation or by taxonomic limitation.
To get a useful answer to your question, you might also need to be more specific. For example:
1. Are you looking for real-time analysis or are you intending to post-process recordings? Real-time analysis could be useful for intelligent triggering of recorders, extending the battery life and reducing unwanted material. Real-time analysis is much harder though than post-processing.
2. Are you looking to implement software on your AudioMoth capture platform or on a downstream system?
3. Is there a particular geographic area of interest?
4. Are there particular taxonomic groups of interest? I know for instance that work has been done characterising the calls of bats, so if bats are your field of interest you may be able to make use of this work.
From what I can see about Arbimon, they appear to be using AI methods in their recognition system, and that is probably the most efficient approach, particularly for systems having low processing power or systems requiring rapid processing of high volumes of data.
If you can't find someone who has already created a recognition system that meets your geographic and/or taxonomic requirements, you may need to consider developing something yourself. Neural network software is readily available. You just need to pre-process the data into forms that are meaningful to your network and then train your network. You may need help from zoological experts to do this.
Good luck with your project.
Very interesting project.
I stumbled upon Preprint Classification of animal sounds in a hyperdiverse rainforest...
There are some other interesting findings too when doing a google search for "training data animal sounds"
- Badges
- Science topic