To provide a speech recognition device and its method for performing a more efficient compensation processing for reducing noise and distortion, etc.
The speech recognition device is constituted by providing a pre-processing part 2 for performing a pre-processing of an acoustic signal, an acoustic parameter calculation means 3 for calculating an acoustic parameter from an output signal of the pre-processing part 2, a matching means 4 for performing a matching processing with a preliminarily registered standard pattern by using the acoustic parameter and for outputting a speech recognition result and an acoustic characteristic estimation means 5 for estimating characteristics added from the acoustic parameter to the acoustic signal and for outputting an estimation result to the pre-processing part 2 and performing compensation by using the estimation result.