Exploring song production mechanisms with an electronic syrinx

J. D. SITT.; E. M. ARNEODO; G. B. MINDLIN; F. GOLLER

Chicago, USA

Conferencia; Neuroscience 2009; 2009

Society For Neuroscience

The basic mechanism of birdsong production resembles the generation of voiced sounds byhumans: expiratory airflow drives sustained oscillations of the membranes - vocal folds in humans and labiain birds. Recently, a modified model (containing nonlinear elastic restitution for the labia) has been proposedto account for a mathematical relationship between fundamental frequency and its spectral content in zebrafinch song. We developed an electronic syrinx that generates song by transducing physiological instructionsinto acoustic output for zebra finches with high similarity in syllable structure between the synthetic andnatural sounds.If syringeal muscles function similarly to those of other songbirds (i.e., ventral syringeal muscles control thetension of the labia, dorsal tracheobronchial muscles affecting gating), we hypothesize that the synergisticaction of the parameters representing the activity of the muscles within the framework of this model allowsgenerating the diversity of complex syllables found in zebra finch song. We used this device to test thishypothesis.The electronic syrinx was implemented using a TMS320C6713 DSK card (spectrum digital), which has aTMS320C6713 Texas Instruments chip. The electronic syrinx acts as an online integrator of the differentialequations of the mathematical models.In order for the model to produce realistic sound with muscle activation patterns we need to switch gain forparts of the song corresponding to high frequency sounds. This suggests that two dynamical regimes are usedin the production of zebra finch song. The parameter controlling the switch is not associated to either,pressure, ventral muscle activity or the activity of the dorsal tracheobronchial muscle.Driving the electronic device implementing our model with the measured physiological instructions weightedby the corresponding gain we perform real time integration of the model equations, allowing us to generatesynthetic song in real time.