Most source separation algorithms are based on a model of stationary source
s. However, it is a simple matter to take advantage of possible nonstationa
rities of the sources to achieve separation. This paper develops novel appr
oaches in this direction based on the principles of maximum likelihood and
minimum mutual information. These principles are exploited by efficient alg
orithms in both the off-line case (via a new joint diagonalization procedur
e) and in the on-line case (via a Newton-like procedure). Some experiments
showing the good performance of our algorithms and evidencing an interestin
g feature of our methods are presented: their ability to achieve a kind of
super-efficiency. The paper concludes with a discussion contrasting separat
ing methods for non-Gaussian and nonstationary models and emphasizing that,
as a matter of fact, "what makes the algorithms work" is-strictly speaking
-not the nonstationarity itself but rather the property that each realizati
on of the source signals has a time-varying envelope.