The Acoustic Adaptation Hypothesis (AAH) and Ecological Character Displacement (ECD) are two potential mechanisms shaping call evolution that can predict opposite trends for the differentiation of signals. Under AAH, signals evolve to minimize environmental degradation and maximize detection against background noise, predicting call homogenization in similar habitats due to environmental constraints on signals. In contrast, ECD predicts greater differences in call traits of closely-related taxa in sympatry because of selection against acoustic interference. We used comparative phylogenetic analyses to test the strength of these two selective mechanisms on the evolution of advertisement calls in glassfrogs, a highly diverse family of neotropical anurans. We found that, overall, acoustic adaptation to the environment may outweigh effects of species interactions. As expected under the AAH, temporal call parameters are correlated with vegetation density, but spectral call parameters had an unexpected inverse correlation with vegetation density, as well as an unexpected correlation with temperature. We detected call convergence among co-occurring species and also across multiple populations from the same species in different glassfrogs communities. Our results indicate that call convergence is common in glassfrogs, likely due to habitat filtering, while character displacement is relatively rare, suggesting that costs of signal similarity among related species may not drive divergent selection in all systems.

We gathered all available glassfrog calls from sound collections, and we supplemented the data set with contributions from colleagues, call descriptions available in literature, and our own recordings obtained in the field. For each species, we obtained two spectral and four temporal call parameters: peak frequency (frequency at which the highest amplitude peak is found, also known as dominant frequency), frequency bandwidth (the difference between upper and lower frequency bounds of notes, as measured 6 dB below the peak frequency), note duration (length in milliseconds of a note, measured from beginning to end of the note), notes per call, pulse rate (number of pulses in a note minus 1, divided by the length of the note; with this formula, purely tonal calls always have a pulse rate of zero) and note rate (count of notes in a call minus 1, divided by the length of the call).

We obtained georeferenced records to infer the distribution for all glassfrog species included in this study from the database generated by Hutter et al. (2013), which was modified with updated species names and supplemented with recent data. For each record, we extracted the environmental temperature, estimated from the mean temperature of the wettest quarter (CHELSA BIO8, Karger and Zimmermann 2019) as a proxy for the temperature during the rainy season, when glassfrog reproductive activity peaks. For H. fleischmanni lineages, we used the available air temperature registered in the metadata of the recording event. We also extracted vegetation density, taken from the Enhanced Vegetation Index (EVI) proposed by the MODIS Land Discipline Group (Liu and Huete 1995) and derived from aerial assessments of forest canopy for both datasets, for this, we obtained the mean value per cell for 204 layers with monthly data between 2000 and 2017.