In a previous work we presented a new architecture dedicated to embedded speech recognition. It relies on a general GMM, which represents the whole acoustic space, associated with a set of HMM state-dependent probability functions modeled as transformations of this GMM. This work takes advantage of this architecture to propose a fast and efficient way to adapt the acoustic models. The adaptation is performed only on the general GMM model and does not require state-dependent adaptation data. It is also very efficient in terms of computational cost. We evaluate our approach in the voice-command task. This adaptation method achieved a relative error-rate decrease of about 10% even if few adaptation data are available .