Back IRMAS: A Dataset for Instrument Recognition in Musical Audio Signals
IRMAS: A Dataset for Instrument Recognition in Musical Audio Signals
We are glad to announce the release of a dataset for Instrument Recognition in Musical Audio Signals (IRMAS dataset).
This dataset was used in the evaluation of the article:
Bosch, J. J., Janer, J., Fuhrmann, F., & Herrera, P. “A Comparison of Sound Segregation Techniques for Predominant Instrument Recognition in Musical Audio Signals”, in Proc. ISMIR (pp. 559-564), 2012
IRMAS is intended to be used for training and testing automatic instrument recognition methods, in a varied set of professionally produced western music recordings. The dataset includes a total of 6705 excerpts for training, and 2874 excerpts for testing. The instruments considered are: cello, clarinet, flute, acoustic guitar, electric guitar, organ, piano, saxophone, trumpet, violin, and human singing voice.
Further information about the music collection, and how the samples were created and annotated is available on the dataset website, where you can also download the audio excerpts and metadata. Given the size of the collection (over 10Gb), you can also first download a sample of the testing and training data, to see if it fits your needs.
We hope that IRMAS will be useful to our scientific community, and we would be very interested in receiving your feedback.