Abstract : When several realizations of an unknown recurrent signal are observed apart from a time expansion or compression, the classical way of estimating these time scaling factors is to take one signal as reference for the estimation. This approach does not take into account the common information between all possible couples of realizations. To achieve this task we use a Maximum-Likelihood based method, in a sub-optimal manner. Using some realistic assumptions and simplifications, we propose a tractable solution. The improvement of classical results is shown through a simulation whose conclusion is that the larger the number of realizations, the more correct the estimation. Finally, we apply the method to electrically evoked sEMG.