This paper focuses on the use of microphone arrays for speaker independent continuous speech recognition in real environment. An array of four omnidirectional microphones was placed at 1.5 m distance from the talker; given the array signals, a Time Delay Compensation (TDC) technique was applied to provide a beamformed signal, that is shown effective as input to a Hidden Markov Model (HMM) based recognizer. The paper also refers to three enhancement/normalization techniques that are employed to obtain a better alignment between the new noisy conditions and the clean ones, under which the training was carried out. In particular, a phone HMM adaptation technique seems to be the most promising for developing a scenario of rapid speaker/environment/ channel adaptation of a realtime hands-free recognizer.