The recent development of advanced speech recognition systems is starting to obtain a serious foothold in health care environments. IBM, Philips and Kurzweil are already designing specific medical application packages, while other companies, like Voice Input Technologies and Voice Activated Systems Technologies, have adapted existing speech recognition software for practical use in Hospital Information Systems (HIS). The physician will soon be able to create a fully computerized patient record (CPR) by simple and direct dictation of the right data into the system for seamless transcription into an online chart. In the June 1998 issue of the Physicians and Computers Journal, Dr. John Leipsic presents a survey of the speech recognition tools, currently available.
The industry selected radiology as the first field to be addressed for speech recognition implementation but other specialities soon followed, such as cardiology, emergency care, hematology, oncology, orthopaedics, pathology, psychiatry, and endocrinology. All these areas require a context specific language model to guarantee optimum speed and accuracy in the process of speech recognition. Voice Automated, an innovative developer, has analysed thousands of medical reports within each speciality to create a model, consisting of 8000 to 12.000 added words to cover 97% to 99% of the vocabulary used in each field. In this way, physicians don't have to waste time on introducing context-specific words to the system.
Already in 1996, the Institute of Medicine stated that the CPR ideally has to support direct physician speech input as well as integrated and interfaced voice dictation via icon on the clinical workstation. Thus, the CPR database offers access for subsequent patient contacts and report generation through algorithms and data mining. According to Dr. John Leipsic, the very first real time continuous speech recognition system for physicians was released by IBM in September 1996. This MedSpeak software package is able to deal with multiple users and includes dictation stages. Philips responded to the challenge with SpeechMagic, an alternative tool featuring a unique batch-processed architecture for report transcription in two stages.
First, the dictation and transcription software SpeechNote records a speech file on Microsoft Windows, after which SpeechMagic converts the file to text. Mobile dictation on the road is enabled by the SpeechPad product. Another feature, also used in IBM's MedSpeak, is the SpeechMike, which consists of a handheld microphone with built in trackball mouse and speaker. The wireless headset microphone is optional but very handy for specialists who prefer to keep their hands free for research and examination purposes. Voice Input Technologies has based its SpeechWriter Mental Health system for group and hospital practices on Philips' SpeechMagic tool with additional integration of a mental health language model.
Microsoft has acquired an 8% interest in the Flemish speech and language technologist Lernout & Hauspie to soundly investigate the incorporation of speech recognition into the Windows operating system, next to handwriting and visual recognition. The Lernout & Hauspie Voice Xpress for medical use is about to be released. Voice Automated has provided the high performance NaturallySpeaking speech recognition software of Dragon Systems with a set of medical speciality language models. The system is suited for current memorization of up to 30.000 frequently used words and stores a vocabulary of more than 230.000 words. Editing and formatting is done at a rate of 150 words a minute with an accuracy between 90% and 95% percent.
The Express Note Taker, developed by Voice Activated Systems Technologies, is based on the Dragon NaturallySpeaking system but features templates covering major diagnoses and standard reporting formats to facilitate the creation of the CPR. In fact, the physician only needs to dictate the patient specific data for transformation into an automatically generated report by means of templates that ignore redundancy in dictation and benefit from the repetitive nature in medical procedures. In case indispensable information has been omitted, the system immediately alerts the physician. Speech recognition software will be integrated into the HIS as a modular component by means of health care specific information exchange protocols, like HL-7. In order to guarantee secure and confidential treatment of sensible patient data, encryption codes and firewalls will be introduced.