HMM adaptation using a phase-sensitive acoustic distortion model for environment-robust speech recognition