Bat Speech Recognition
Problem being addressed
Microchiroptera, more commonly known as bats, have unique voices with interesting acoustic signatures. How can different calls, from different species, be told apart?
Two well known speech recognition machine learning techniques: Gaussian Mixture Modelling (GMM) and Hidden Markov Modelling (HMM) are applied to a labelled data set of bat calls that originate from different species of bat. The calls are specifically search phase calls that are used in echo-location (by the bat) and the training data has been hand labelled by experts. The GMM model, in particular, was found to be extremely accurate in predicting the species from which the batcall originated.
Advantages of this solution
Adapting tehcniques that are essentially developed for humans such as speech recognition to other species is an interesting area of research that combines zoology and computer science. The fact that we can accurately predict and identify species in the wild with implementations of these types of models in real-time will be of great benefit to wildlife enthusiasts and conservationists.
Solution originally applied in these industries
Possible New Application of the Work
Electronics and Sensors Industry
Noise cancelling devices could perhaps benefit from solving the reverse of this problem which would be to create the inverse of a sound wave (using AI) in order to damp out sounds from the outside.
This research could be reapplied to the engineering of sonar and other sound based location systems. The basic idea of finding spatial information through sound waves comes from bat calls, and understanding these calls could perhaps be useful in designing custom sonar systems.
In some types of mining operations are carried in hazardous environments deep underground. Using an AI that is backed by lessons from nature, such as bat calls for depth perception to "see" terrain underground could be useful in the mining industry.
Source DOI: #############
Source URL: #############