Open AccessEnhancing MESSL algorithm with robust clustering based on Student's t ‐distribution
Author(s) -
Zohny Z.Y.,
Naqvi S.M.,
Chambers J.A.
Publication year - 2014
Publication title -
electronics letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.375
H-Index - 146
ISSN - 1350-911X
DOI - 10.1049/el.2013.4230
Subject(s) - cluster analysis , outlier , algorithm , univariate , parametric statistics , computer science , mixture model , probabilistic logic , student's t distribution , gaussian , mathematics , speech recognition , pattern recognition (psychology) , artificial intelligence , statistics , machine learning , multivariate statistics , volatility (finance) , physics , quantum mechanics , econometrics , autoregressive conditional heteroskedasticity
The model‐based expectation maximisation source separation and localisation (MESSL) algorithm is enhanced through the integration of robust clustering based on the Student's t ‐distribution. This heavy‐tailed distribution, as compared with the Gaussian distribution used in MESSL, can potentially capture in a better manner the outlier values in the univariate parametric modelling of the time–frequency (T–F) points and thereby lead to more accurate probabilistic masks for source separation. In particular, the Student's t ‐distribution is exploited in modelling the interaural phase difference (IPD) in order to represent in a better manner the uncertainties introduced by the statistical non‐stationarity of the speech signals and the associated small sample length effects. Simulation studies based on speech mixtures formed from the TIMIT database confirm the advantage of the proposed approach in terms of the signal to distortion ratio (SDR).