The results of four sequential experiments were combined into integrated empirical models using data bridging. The resulting regression models can be viewed as quantitative design guidelines for telephone-based information systems. A total of ten independent variables involving environmental, hardware, dialogue, and user factors were considered across the four sequential studies that evaluated a telephone-based interface. Three dependent variables including total search time, user added keypresses, and message transcription accuracy were evaluated in each separate study. Polynomial regression was used to generate an integrated second-order empirical model for each of the three dependent variables. The major contributors to total search time were the time delay between the presentation of each menu item (input timeout) and the structure of the menus. Age of the user and menu structure were the primary contributors to user added keypresses required to recover from errors. Overall, the accuracy of message transcription was influenced primarily by background music, presentation rate of the synthesized speech, and the age of the user. Total search time in this sequential type of information delivery system is primarily dialogue driven. Critical dialogue parameters for this system were input timeout and the number of alternatives in menus. Due to the small number of errors in searching for information, the need to minimize user added keypresses for error correction was not of primary design importance. Accuracy of speech message transcriptions was primarily dependent upon the acoustical environment of the listener. Overall, the use of integrated empirical models offers several advantages including a basis for generalization across several studies and the ability to conduct specific interface design tradeoffs.
INTRODUCTIONTelephone information systems can be used to convey simple serial information. In these systems the user communicates with a computer using a basic telephone to access voice mail, to search a database, or to alter a database. Because of the limited data entry capability of the telephone keypad, most successful applications have emphasized data delivery rather than data entry. Automatic speech recognition may provide a useful method of telephone data entry in the future, and many researchers are working toward video telephones. However, the present research focused on the design of information systems using the standard telephone because this will be the primary means of access in the home and in small businesses for many years. It should be noted that synthesized rather than recombinant speech was used in this research. Because speech intelligibility is generally lower for synthesized speech, some of these results may not apply to systems using recombinant speech.To design effective telephone based information systems it is useful to have empirical design guidelines relating to hardware, user, dialogue, and environmental considerations. Meister (1985) discusses the need to provide design guidelines tha...