A toolkit for matrix factorization of soundscape spectrograms into independent streams of sound objects, possibly representing individual species or independent group behaviours.

The method employs shift-invariant probabilistic latent component analysis (SIPLCA) for factoring a time-frequency matrix (2D array) into a convolution of 2D kernels (patches) with sparse activation functions.

Methods are based on the following:

1. Smaragdis, P, B. Raj, and M.V. Shashanka, 2008. Sparse and shift-invariant feature extraction from non-negative data. In proceedings IEEE International Conference on Audio and Speech Signal Processing, Las Vegas, Nevada, USA.

2. Smaragdis, P. and Raj, B. 2007. Shift-Invariant Probabilistic Latent Component Analysis, tech report, MERL technical report, Camrbidge, MA.

3. A. C. Eldridge, M. Casey, P. Moscoso, and M. Peck (2015) A New Method for Ecoacoustics? Toward the Extraction and Evaluation of Ecologically-Meaningful Sound Objects using Sparse Coding Methods. PeerJ PrePrints, 3(e1855) 1407v2 [In Review]

Either run python (or jupyter notebook) in the installed SoundscapeEcology directory, or add the directory's path to your $PYTHONPATH environment variable.

To test that everything is installed correctly, launch a python shell or notebook and type:

from bregman.suite import *
from soundscapeecology import *

Please report any problems in the issue tracker.