More about hybrid-vocal-classifier and songbird science¶
more about hybrid-vocal-classifier¶
Scientists that study how birdsong at the level of individual birds often have to label the syllables of each bird’s song by hand in order to get results. Because birds can sing hundreds of songs a day, it requires many person-hours to label song. In addition, in many songbird species that have been studied, each individual learns a song that is similar but not exactly the same as the song of the bird or birds that tutored it. Therefore any software that automates the process of labeling syllables must classify them with very high accuracy and must do so in a way that is robust across the songs of many different individuals from a species.
The primary goal of hybrid-vocal-classifier is to make it easier for any scientist to apply machine-learning algorithms to birdsong.
The secondary goal of the package is to facilitate comparisons of different machine learning algorithms. Several groups have published on various algorithms: for a list of some of the algorithms used by this library, see Citations, and see related works in Related articles and software. Along with the library, a large dataset containing days of hand-labeled song was released. Open repositories were crucial for other advances in machine learning. A list with links to that dataset and others can be found on the Data repositories page.
A final goal is to further develop the links between machine learning, artificial intelligence, and the study of birdsong. Obviously, the work most related to this library is in automated speech recognition. Birdsong presents an ideal test-bed to experiment with related machine learning algorithms, e.g., for segmenting time-series data.
more about songbird science¶
Songbirds provide a model system to understand how the brain learns and produces speech and other sequential motor skills acquired by imitation, like playing the piano or shooting a basketball. Like babies learning to talk, songbirds learn their song from an adult “tutor” during a sensitive period in development . Each individual bird has its own unique song, often very similar to the song of the bird that tutored it. The songbird brains contain a specialized network of areas required for learning and producing song. Although this network, known as the song system, is found only in songbird brains, it has evolved on top of the basic floor plan that appears in all vertebrate brains, including humans. By understanding the song system, we can understand more about our own brain. Many other aspects of songbird behavior can tell us more about ourselves and our environment, and by studying their vocalizations we open a window into their world. For more information, check out http://songbirdscience.com/
hybrid-vocal-classifier was developed in the Sober lab