A maximum likelihood approach to single-channel source separation

Authors:
Gil-Jin Jang;Te-Won Lee
Affiliations:
Spoken Language Laboratory, Division of Computer Science, KAIST, Daejon 305-701, South Korea;Institute for Neural Computation, University of California, San Diego, La Jolla, CA
Venue:
The Journal of Machine Learning Research
Year:
2003

Citing 11
Cited 3

What is the goal of sensory coding?

Neural Computation
Independent component analysis, a new concept?

Signal Processing - Special issue on higher order statistics
An information-maximization approach to blind separation and blind deconvolution

Neural Computation
Listening to two simultaneous speeches

Speech Communication
Sparse code shrinkage: denoising of nongaussian data by maximum likelihood estimation

Neural Computation
Blind Source Separation by Sparse Decomposition in a Signal Dictionary

Neural Computation
Learning Overcomplete Representations

Neural Computation
Single channel separation using linear time varying filters: separability of non-stationary stochastic signals

ICASSP '99 Proceedings of the Acoustics, Speech, and Signal Processing, 1999. on 1999 IEEE International Conference - Volume 03
Blind separation of mixture of independent sources through aquasi-maximum likelihood approach

IEEE Transactions on Signal Processing
Blind source separation-semiparametric statistical approach

IEEE Transactions on Signal Processing
Separation of speech from interfering sounds based on oscillatory correlation

IEEE Transactions on Neural Networks

Indexing and matching of polyphonic songs for query-by-singing system

Proceedings of the 12th annual ACM international conference on Multimedia
The detection of changes of the auditory scene structure using a concept of short-time ICA

ICCOM'05 Proceedings of the 9th WSEAS International Conference on Communications
Single channel audio source separation

WSEAS Transactions on Signal Processing

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper presents a new technique for achieving blind signalseparation when given only a single channel recording. The mainconcept is based on exploiting a priori sets of time-domainbasis functions learned by independent component analysis (ICA) tothe separation of mixed source signals observed in a singlechannel. The inherent time structure of sound sources is reflectedin the ICA basis functions, which encode the sources in astatistically efficient manner. We derive a learning algorithmusing a maximum likelihood approach given the observed singlechannel data and sets of basis functions. For each time point weinfer the source parameters and their contribution factors. Thisinference is possible due to prior knowledge of the basis functionsand the associated coefficient densities. A flexible model fordensity estimation allows accurate modeling of the observation andour experimental results exhibit a high level of separationperformance for simulated mixtures as well as real environmentrecordings employing mixtures of two different sources.