Constructional details of speech recognition systems