A Jungk1, B Thull, L Fehrle, A Hoeft, G Rau. 1. Helmholtz Institute for Biomedical Engineering, Aachen University of Technology, Aachen, Germany. jungk@hia.rwth-aachen.de
Abstract
OBJECTIVE: New developments in speech interaction technology warrant the assumption that some of the interaction problems at anesthesia workplaces can be solved using speech interaction. One application might be the documentation of the anesthetic procedure. METHOD: To investigate this hypothesis, a speech input system was developed for a commercially available patient monitor (Datex-Ohmeda AS/3 Anesthesia Monitor with Record-Keeper), which in addition to standard functions also facilitates the generation of the anesthesia record. The speech input system was based on the IBM ViaVoice speech recognition technology (version Millennium Pro). The speech input device was preliminarily tested by 5 anesthetists wearing surgical masks under laboratory conditions and through retrospectively documenting 4 to 5 anesthetic procedures (each 12 to 17 documentation entries) in a calm situation. The speech interactive documentation procedure was compared with the conventional keyboard documentation procedure. RESULTS: The test persons achieved an overall word recognition rate between 72% and 92.4%. Between 76.5% and 97.7% of the single word commands could be completed without any corrections. The same qualifier held for 58.5% to 86.3% of the two- and three-word commands. Despite high speech recognition error rates, almost all documentation entries (236 of 238) could be finished using speech interaction. Comparing both documentation techniques, speech interaction was said to be more intuitive and caused less frustration than the keyboard documentation procedure. However, speech interactive documentation was rated less controllable because of the high speech recognition error rates and bccause the auditive feedback dialogues took too long. CONCLUSIONS: Modern speech recognition tools are still not advanced enough to facilitate the design of applications with an almost natural speech interface and widespread user acceptance. Nevertheless, many tasks in anesthesia have the necessary characteristics to be optimally supported by speech interaction. In contrast to earlier approaches to speech-interactive anesthesia workplaces, successful application today depends on the question of design rather than solely on that of technology. Many of the constraints and drawbacks of current technology can be overcome through appropriate design measures. The goals must focus first on identifying task areas in intensive care where speech-interaction can yield real benefit in terms of work efficiency, and second on developing and evaluating an ergonomic design of speech interaction. The intended users seem to look forward to the incorporation of speech interaction at the workplace.
OBJECTIVE: New developments in speech interaction technology warrant the assumption that some of the interaction problems at anesthesia workplaces can be solved using speech interaction. One application might be the documentation of the anesthetic procedure. METHOD: To investigate this hypothesis, a speech input system was developed for a commercially available patient monitor (Datex-Ohmeda AS/3 Anesthesia Monitor with Record-Keeper), which in addition to standard functions also facilitates the generation of the anesthesia record. The speech input system was based on the IBM ViaVoice speech recognition technology (version Millennium Pro). The speech input device was preliminarily tested by 5 anesthetists wearing surgical masks under laboratory conditions and through retrospectively documenting 4 to 5 anesthetic procedures (each 12 to 17 documentation entries) in a calm situation. The speech interactive documentation procedure was compared with the conventional keyboard documentation procedure. RESULTS: The test persons achieved an overall word recognition rate between 72% and 92.4%. Between 76.5% and 97.7% of the single word commands could be completed without any corrections. The same qualifier held for 58.5% to 86.3% of the two- and three-word commands. Despite high speech recognition error rates, almost all documentation entries (236 of 238) could be finished using speech interaction. Comparing both documentation techniques, speech interaction was said to be more intuitive and caused less frustration than the keyboard documentation procedure. However, speech interactive documentation was rated less controllable because of the high speech recognition error rates and bccause the auditive feedback dialogues took too long. CONCLUSIONS: Modern speech recognition tools are still not advanced enough to facilitate the design of applications with an almost natural speech interface and widespread user acceptance. Nevertheless, many tasks in anesthesia have the necessary characteristics to be optimally supported by speech interaction. In contrast to earlier approaches to speech-interactive anesthesia workplaces, successful application today depends on the question of design rather than solely on that of technology. Many of the constraints and drawbacks of current technology can be overcome through appropriate design measures. The goals must focus first on identifying task areas in intensive care where speech-interaction can yield real benefit in terms of work efficiency, and second on developing and evaluating an ergonomic design of speech interaction. The intended users seem to look forward to the incorporation of speech interaction at the workplace.