Memory corruption during voice activation, when sound model parameters are loaded from HLOS to ADSP.