Songbirds provide a powerful model system for studying sensory-motor learning. Many analyses require time-consuming manual annotation of the units of song, often called syllables. However, songbirds produce many more songs than can be annotated by hand, creating a bottleneck that limits the questions researchers can answer. Methods exist for automated annotation[1], but, as we show, these methods exhibit limitations when applied to songs with many syllable types and variable transitions. To address these issues, we developed an artificial neural network, TweetyNet[2], that segments and labels spectrograms. We first show how our approach mitigates issues faced by methods that rely on segmented audio. Then we demonstrate TweetyNet achieves low error across individuals and across two species, Benaglese finches and canaries. We also show that using TweetyNet, we can accurately annotate very large datasets, containing song recorded in multiple days, and that these predicted annotations replicate key findings from behavioral studies in both species[3,4]. TweetyNet is the first algorithm to automate annotation of canary song at the syllable level, processing minutes-long bouts with as many as 50 syllable types. We demonstrate that access to thousands of songs yields an order of magnitude precision improvement in statistical syntax models and allows measurements of previously-unknown diurnal changes. Specifically, probabilities of different first-order Markov sequences exhibit uncorrelated changes. This observation suggests that beyond global parameters, like temperature and neuromodulator tone, there may be additional mechanisms that affect canary syntax. We then estimate the diurnal change in transition entropy and find decrease in second-order but not in first-order Markov processes. These differences indicate dynamics in the long-range structure of song, similar in trend to the known diurnal decrease in syllable acoustic variability, but on tenfold longer time scales. Our results suggest TweetyNet will make it possible to address a wide range of new questions about birdsong.