/**
* Outlier elimination algorithms for voice conversion.
* The aim of outlier elimination is to detect source and target pairs
* that might result in reduced conversion performance or artefacts.
* The best working method so far has been the GaussianOutlierEliminator.
* This method fits a single Gaussian to the difference distributions of various
* source and target acoustic features (LSFs, duration, f0, energy).
* Then, the pairs with significant difference as compared to distribution mean are eliminated.
* The voice conversion training step can also be set to eliminate too close pairs, forcing
* an average amount of unlikeliness in the training data.
*/
package marytts.signalproc.adaptation.outlier;