The embodiments of the present disclosure provide a device for recognizing speeches and a method for speech recognition. The device for recognizing speeches may comprise a processor, configured to execute instructions stored in the memory, to: perform speech recognition on the collected audio data to obtain a semantic content of the audio data; match the obtained semantic content with a semantic data stored in the database; determine whether the audio data contains ambient noise audio information and audio information of a user, in response to determining that the obtained semantic content does not match with the semantic data; and change conditions for collecting the audio data and control to collect the audio data with the changed conditions, in response to determining that the audio data contains the ambient noise audio information and the audio information of the user.