To provide a speech recognition system which eliminates the need for a user to generate a sound model repeatedly and can recognize speech with high precision in itself even when a plurality of speech recognition systems desired by the user are present.

The speech recognition system 1 comprises following components. A sound model generating device 50 generates a sound model 51 for speech recognition targeting the user as a specified speaker. A mobile terminal device 2 stores the sound model 51 acquired from the sound model generating device 50, and outputs the sound model 51 to a speech recognition device 100 to recognize the speech of the specified speaker. The speech recognition device 100 comprises a recognition-device side sound model acquiring means 6 of acquiring the sound model 51 from the mobile terminal device 2, a sound model registering means 5F of registering the acquired sound model 51, and a speech recognition means of recognizing and specifying recognized object information included in an input speech wave on the basis of information comparison beteen the speech wave and sound model 51.