Description: 
If you go to Amazon.com or the Apple Itunes store, your ability to search for new music will largely be limited by the `query-by-metadata' paradigm: search by song, artist or album name. However, when we talk or write about music, we use a rich vocabulary of semantic concepts to convey our listening experience. If we can model a relationship between these concepts and the audio content, then we can produce a more flexible music search engine based on a 'query-by-semantic- description' paradigm. In this talk, I will present a computer audition system that can both annotate novel audio tracks with semantically meaningful words and retrieve relevant tracks from a database of unlabeled audio content given a text-base query. I consider the related tasks of content- based audio annotation and retrieval as one supervised multi-class, multi-label problem in which we model the joint probability of acoustic features and words. For each word in a vocabulary, we use an annotated corpus of songs to train a Gaussian mixture model (GMM) over an audio feature space. We estimate the parameters of the model using the weighted mixture hierarchies Expectation Maximization algorithm. This algorithm is more scalable to large data sets and produces better density estimates than standard parameter estimation techniques. The quality of the music annotations produced by our system is comparable with the performance of humans on the same task. Our `query-by-semantic-description' system can retrieve appropriate songs for a large number of musically relevant words. I also show that our audition system is general by learning a model that can annotate and retrieve sound effects. Lastly, I will discuss three techniques for collecting the semantic annotations of music that are needed to train such a computer audition system. They include text-mining web documents, conducting surveys, and deploying human computation games.

Runtime:1:01:12

Rating:TV-G