A voice based multimodal speaker authentication method and telecommunications application thereof employing a speaker adaptive method for training phenome specific Gaussian mixture models. Applied to telecommunications services, the method may advantageously be implemented in contemporary wireless terminals.