Generation and Recognition of Bird Songs

Recent experimental findings in songbirds shed some light onto the mechanisms underlying birdsong generation. The vocal pathway of songbirds consists of HVC (High Vocal Center), RA nucleus and respiratory and syringeal areas. The syringeal areas control the avian vocal organ, the syrinx. This hierarchical organization of regions is one of the most studied areas in songbirds. The relation between HVC and RA regions was recently revealed by a study [1] which showed the neurons in HVC that project onto RA, fire sparsely at a single, precise time during the song and drive a subpopulation of RA neurons (see the Figure) in the lower level. RA neurons driven by HVC control the respiratory and the syringeal muscles which are the final steps of producing complex songs.

Summary of the birdsong generation model: The sequential activation of HVC neurons drives a subpopulation of RA neurons in the lower level. RA neurons innervate the syringeal and respiratory areas which produce the final output.

In this project, we use dynamical systems theory to model this hierarchical organization starting from HVC. The neurons in HVC activate RA neurons by creating attractors in the lower level during their activation time. We then use this activation of RA neurons as control parameters for a mathematical model of the syrinx. The output is a complex, synthetic birdsong.
After generation, we try to understand how a second bird could recognize this song. This is an example of an inverse problem and we use novel variational Bayesian techniques to invert the generation scheme. The inversion means that the second bird can infer the states of HVC and RA neurons which generated this song. We test our model and inversion scheme with several simulations that also give insight into communication of birds within species.

 

References

[1] An ultra-sparse code underlies the generation of neural sequences in a songbird, R.H.R. Hahnloser, A.A. Kozhevnikov, M.S. Fee , Nature,419: 65-70 (2002).
First results of this project have been published in:
[2] Yildiz IB , Kiebel SJ , 2011 A Hierarchical Neuronal Model for Generation and Online Recognition of Birdsongs. PLoS Comput Biol 7(12): e1002303. doi
[3] Yildiz IB, von Kriegstein K, Kiebel SJ, 2013, From birdsong to human speech recognition: Bayesian inference on a hierarchy of nonlinear dynamical systems. PLoS Comput Biol, 9(9): e1003219 doi

Go to Editor View