Many animals produce distinct sounds or substrate-borne vibrations, but these signals have proved challenging to segment with automated algorithms. We have developed SongExplorer, a web-browser based interface wrapped around a deep-learning algorithm that supports an interactive workflow for (1) discovery of animal sounds, (2) manual annotation, (3) supervised training of a deep convolutional neural network, and (4) automated segmentation of recordings. Raw data can be explored by simultaneously examining song events, both individually and in the context of the entire recording, watching synced video, and listening to song. We provide a simple way to visualize many song events from large datasets within an interactive low-dimensional visualization, which facilitates detection and correction of incorrectly labelled song events. The machine learning model we implemented displays higher accuracy than existing heuristic algorithms and similar accuracy as two expert human annotators. We show that SongExplorer allows rapid detection of all song types from new species and of novel song types in previously well-studied species.

11The neural basis for behavioural evolution is poorly understood. Functional comparisons 12 of homologous neurons may reveal how neural circuitry contributes to behavioural 13 evolution, but homologous neurons cannot be identified and manipulated in most taxa.14 Here, we compare the function of homologous courtship song neurons by exporting 15 neurogenetic reagents that label identified neurons in Drosophila melanogaster to D. 16 yakuba. We found a conserved role for a cluster of brain neurons that establish a 17 persistent courtship state. In contrast, a descending neuron with conserved 18 electrophysiological properties drives different song types in each species. Our results 19suggest that song evolved, in part, due to changes in the neural circuitry downstream of 20 this descending neuron. This experimental approach can be generalized to other neural 21 circuits and therefore provides an experimental framework for studying how the nervous 22 system has evolved to generate behavioural diversity. 24peer-reviewed) is the author/funder. All rights reserved. No reuse allowed without permission.The copyright holder for this preprint (which was not . http://dx.doi.org/10.1101/238147 doi: bioRxiv preprint first posted online Dec. 21, 2017; 2 however, do not sing sine song, but they produce two distinct modes of pulse song: thud 54 peer-reviewed) is the author/funder. All rights reserved. No reuse allowed without permission. Main textThe copyright holder for this preprint (which was not . http://dx.doi.org/10.1101/238147 doi: bioRxiv preprint first posted online Dec. 21, 2017; 3 song and clack song 11 . Thud song is generated by unilateral wing vibration; while clack 55 song is generated when males vibrate both wings behind them (Fig. 1a) moving faster than when males sing pulse song (Fig. 1e, f). Additionally, males sing 68 pulse song mostly when they are located directly behind females, whilst they sing clack 69 song across a wide range of distances and positions relative to females (Fig. 1g) 11 . 70Consistent with these observations, removing motion signals by providing males with a 71 motionless decapitated female eliminated clack song but not pulse song (Fig. 1h, i shown to reduce wing extension during courtship 15 . We found that pIP10 inhibition in D.144 peer-reviewed) is the author/funder. All rights reserved. No reuse allowed without permission.The copyright holder for this preprint (which was not . http://dx.doi.org/10.1101/238147 doi: bioRxiv preprint first posted online Dec. 21, 2017; 6 melanogaster using our new split-GAL4 line caused almost complete elimination of 145 pulse song and a small reduction in sine song produced during normal courtship (Fig. 146 2a). In contrast, pIP10 inhibition in D. yakuba eliminated clack song consistently across 147 different split-GAL4 drivers and neuronal inhibitors (Fig. 2b and Extended Data Fig. 5a-148 c). In addition, in some treatments, pIP10 inhibition resulted in a quantitative reduction 149 of pulse song (Extended Data Fig. 5a, c). Th...