Introduction: Design of speech-based controls refers to systems that recognize human speech and treat speech commands as inputs to the ATIS system. As discussed in Reference 1, automatic speech recognition (ASR) systems may be characterized with respect to three sets of design characteristics. First, speaker-dependent systems recognize speech from only one speaker that has been calibrated to the system; speaker-independent systems can recognize speech from many speakers. Second, isolated word recognition systems require that speakers provide a pause or gap between words in a message; continuous speech recognition systems do not require any pause between words. Third, ASR systems vary with respect to the size of the vocabulary that they recognize.

Issues to Consider When Designing ASR Systems

Supporting Rationale: Reference 2 provides considerable discussion of issues and research related to speech controls; the guidelines presented above have been adapted from design principles presented in Reference 2 and, to a lesser extent, Reference 1. The guidelines presented above reflect limited experience in the use of speech as a control device from two technical domains: (1) military information systems and flight control, and (2) the telecommunications field. Case studies and anecdotal results from several applications of speechcontrols can be found in References 1 and 2. Although various commercial speech recognition systems have been developed for automotive applications, published empirical results are few and have not always provided consistent design guidance.

Special Design Considerations: As noted in Reference 2, key issues in the design and implementation of ASR systems include:

• Recognition accuracy: Lower accuracies will reduce system performance and user acceptance.

• Background noise: Ambient noise (traffic, radio, speech displays) can interfere with ASR system performance.

• Speech variability: Human speech varies considerably with respect to volume, frequency, pitch, and tone under different conditions. Speech variability can contribute to reduced recognition of speech.

• Task selection: Selection of tasks for which speech should be used must reflect task characteristics and a clear understanding of the trade-offs associated with using speech controls vs. manual controls.