This paper describes a framework for distributed speech recognition (DSR) that exploits the information about the acoustic environment in order to improve the system per- formance. At the client side, the ETSI front-end is used. At the server, we propose an acoustic ambient detector that identifies the acoustic environment so that the recognition engine can employ a set of acoustic models which match the identified environment. The experimental results show that the proposed framework outperforms a system using the ETSI advanced front-end (AFE) with multicondition-trained acoustic models both in well-matched and mismatched conditions. Furthermore, our proposal maintains a more light-weighted front-end, which enables DSR in resource- limited clients (cellular phones or PDAs).