![]() Many models and analyses rely on song annotated at the syllable level, including: statistical models of syntax ( Markowitz et al., 2013 Jin et al., 2011 Berwick et al., 2011 Hedley, 2016) computational models of motor learning ( Sober and Brainard, 2009 Sober and Brainard, 2012) and analyses that relate both acoustic features and sequencing of syllables to neural activity ( Leonardo and Fee, 2005 Sober et al., 2008 Wohlgemuth et al., 2010). Labels correspond to a set of discrete syllable classes that a researcher defines for each individual bird. First, they segment song into units, often called syllables, and second, they assign each syllable a label. To annotate birdsong, researchers follow a two-step process ( Thompson et al., 2012 Kershenbaum et al., 2016). Annotation is a time-consuming process done by hand with graphical user interface (GUI) applications, for example Praat, Audacity, Chipper ( Boersma and Weenink, 2021 Audacity Team, 2019 Searfoss et al., 2020). However, similarly scaling up other analyses of vocal behavior is currently hindered by a lack of automated methods.Ī major roadblock to scaling up many analyses is that they require researchers to annotate song. For example, automated methods for measuring similarity of juvenile and tutor song across development ( Tchernichovski et al., 2000 Mets and Brainard, 2018a) led to important advances in understanding the behavioral and genetic bases of how vocalizations are learned ( Tchernichovski et al., 2001 Mets and Brainard, 2018b Mets and Brainard, 2019). Leveraging this amount of data requires methods for high-throughput automated analyses. Their natural behavior yields a detailed readout of how learned vocalizations are acquired during development and maintained in adulthood. A key advantage of songbirds as a model system is that birds sing spontaneously, producing hundreds of song bouts a day. In this and many other ways, birdsong resembles speech ( Brainard and Doupe, 2002). Their songs consist of vocal gestures executed in sequence ( Fee and Scharff, 2010). Juveniles typically learn song from a tutor, like babies learning to talk. Birdsong is a culturally transmitted behavior learned by imitation ( Mooney, 2009). ![]() ![]() Songbirds are an excellent model system for investigating sensory-motor learning and production of sequential behavior. We conclude that TweetyNet makes it possible to address a wide range of new questions about birdsong. In addition, we provide open-source software to assist other researchers, and a large dataset of annotated canary song that can serve as a benchmark. Lastly, we demonstrate that using TweetyNet we can accurately annotate very large datasets containing multiple days of song, and that these predicted annotations replicate key findings from behavioral studies. ![]() We also show that TweetyNet performs well across multiple individuals from two species of songbirds, Bengalese finches and canaries. ![]() We show that TweetyNet mitigates limitations of methods that rely on segmented audio. Here, we present TweetyNet: a single neural network model that learns how to segment spectrograms of birdsong into annotated syllables. Automated methods for annotation have been proposed, but these methods assume that audio can be cleanly segmented into syllables, or they require carefully tuning multiple statistical models. However, many analyses of birdsong require time-consuming, manual annotation of its elements, called syllables. Songbirds provide a powerful model system for studying sensory-motor learning. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |