Continuous Speech Recognition for Clinicians - Term Paper Example

Comments on “Continuous Speech Recognition for clinicians.” Abstract: Technology has been infused in the lives of people and all academia are now relying heavily for further advancements and research based in the functions of technology. The amalgamation of the computerised information systems is not just a separate entity in today’s world but is rather a fundamental tool for growth and globalisation. Clinicians have found the same use for technology and many dynamic changes have been made to adopt technology in the form of speech recognition systems for the patients as well as the hospital staff. These technological influences have been known to affect the workload and the nullification of repetitive tasks. The paper goes over some of the aspects in research and how improvements may be made in this framework. Introduction: The authors have researched on the improvements in the voice recognition software and applying these developments in the information systems in the field of medicine. Zafar et al (1999) have extensively discussed the application of such software into the tedious and time consuming repetitive process of transcriptions and handling of medical records. The authors have noted in their research that a single hospital may spend over five hundred thousand dollars per year on medical transcriptions alone. The developed systems of voice and speech recognition can cut this cost significantly and be time efficient as well. As earlier versions of this software was a mere hurdle in the practical use of the product as a substitute for form based applications and typing or note taking tedious inputs- the newer releases have been consistent in recognising voice patterns and understanding the normal speed of speech from humans to take dictation and translate that as input into the information system in the medical field. The system has been improved in these voice recognition software as more research has been done in the studies of sounds and specifically in the types of different phoneme that branch out to make words. These phonemes have been used as the basis of reproduction of text words that the software can match in its built in dictionary or other plausible words. Since a word has a combination of phonemes associated to it and there are phonemes that may either correspond to one alphabet or a group of alphabets in a single word- that makes the basis of the text recognition software in this time. The research presented by the authors has also been extensively studied by many other scholars individually and in papers and seminars as well. Waibel & Lee (1990), Saito and Shuzo (1991), Ayuso and Soler (1995) have based the research upon the fact that human voice comprises of both words and sound that holds no conversational meaning. The sounds other than they words may be categorised as noise as when speaking the human system may also contribute in sounds that may be related to breathing, puffing of air during speaking and other related aspects of the functioning of the vocal cord in the human body that may send out noise along with words. Furthermore, Carter (2001) and Sanfeliu and Schulcloper (2004) have noted that the latest developments in the software have been to differentiate between the sound and the noise patterns based on the waveforms that have been researched to relate specific types to phonemes and noise. Furthermore, the power spectrum of these software’s speech recognition engines have been further attuned to function on the Formants and the Consonants of the voice pattern sin the words. These findings have been the building blocks of the developments in the speech recognition software. Method: The research has been conducted through the practical testing of the voice recognition software in consolidates environments where many aspects of the natural environment or the complexities of it may have been ignored. The voice recognition systems were evaluated on the basis of commercially available software in the market, which have options to add on different dictionaries related to the medical field. These tests were developed to narrate the possibility of the different brands of software being similar in performance and five different packages were picked to test. The software were tested on the system requirements of the vendors that the platforms were selected to match the specifications as per each individual brand of speech recognition software. A comparison of these software were done in newer editions versus the older ones as well apart from just comparing over different brands. Two trials were conducted in the methodology of the comparison where each trial differed in the system the speech recognition tool was being used on and the variance in the version in relation to the diverse base of the dictionaries that were used. The software was tested by dictating fifty discharge summaries in the first trial and it was observed that the software did not pick up about 400 words of the dictation that was given to it. The software had mainly not found the se words in the dictionary that was used and in many cases had replaced a similar word in the dictation that it gave as an output. Furthermore, many of the mistakes were not due to the absence of the words in the dictionary but words were still mismatched in terms of tense or other similarities in the words that confused the software. The software was further tested and developed for three weeks and only after this period the system showed some practicality in recognising speech and giving appropriate output. However, the system was still off by two percent and showed errors in dictation. The errors still persisted in abbreviations and it was a failure to try to differentiate or train the tool to recognise similar sounding abbreviations. On the front of the system speed it was found that a much faster system was required to make an actual use of the software and the recommended computer would be too slow to process the dictation and would fall behind. For the first set of tests, the system may be found as a lot tedious and complicated to use. It would require an initial period of at least three weeks along with the manual addition of words in its database. This may not be possible an a practical approach in the medical field as the time needed to develop the system for some functionality may not be available and the medical staff using this software may not be that technical to use and develop it accordingly. As some software tested out to be below average in the performance, in the second trial test a different range of software tools were tested and it was found that these software gave excellent results and do not even need the long three week development period in terms of adapting to the voice patterns of the user. The more successful of the second test gave the views that the software with a more developed engine to recognise sound and a better word dictionary in the support of recognising words performed much more consistently. Even though the recommended computer speed had been again a failure in keeping up with the higher speed of dictation but overall the tuning was only needed in terms of the microphone that was used and the initial setting of the software. The software tested in the later stage also did not just work well in the quite office surrounding bit also in the busy and noisy areas of wards and other offices. Research Plan For Further Work: The research by the authors may be extended and refined based on the findings that there are many available software engines that have been practical in the use of taking dictation. The methodology however, may be modified to include a more differentiation in voice patterns by including more people in giving dictation and differentiating between male and female genders as a source of giving dictation (Kiel, 2000). The difference in the speech format in genders may be a more interest in the research as the staff in the hospitals and clinics is already diversified and different waveforms of the vocal chord in the genders may have a substantive effect on the results of the findings. Furthermore, it may also be imperative to include another diversification through the age (Maheu et al, 2004). Subjects giving dictation may be spread over different age groups and the test of the engine may be thoroughly set to see if the voice recognition engine functions in the same way across the age groups. It may be of high interest to find out that the engine may respond well to the clear and crisp voice patterns of a twenty year old or a thirty year old but may not function in the same successful manner for a forty or a fifty year old. Voice recognition technology is based on a speech pattern recognition format and any variation in the speech format may be of the point of interest to be infused in the development and testing of the voice recognition engine (Ranchhod, 2002). Another important aspect of the research can be to compare the results over different dialects of the subjects giving dictation. A specific language such as English may have multiple dialects in which people may communicate with each other. Diverse cultural and regional backgrounds may influence the wave patterns if the voice in subjects and the results of the software’s functionality may be different. The engine may not be able to understand the same words from a typical Texan accent of English where it may have shown success in a normal American accent (Ribeiro et al, 2005). Conclusions: Voice recognition software has had its implication out of the computers and now can be found in many banking operations, cell phones, automated customer service operations for people. The voice recognition systems have evolved out of their primary speech engines and have the ability to recognize pre determined information that is fed in their database to be compared and matched accordingly. However, even though the developments in the industry have been thorough and consistent, there is still more in the development and advancement of the speech recognition software. The speech recognition software need to be tailored to adapt more easily towards different voice patterns and to develop as an intelligent self supportive system in to enrich the functionality of their use. It may be of a plausible research topic that intelligent voice recognition systems may be developed which may be pre-configured for the adaptation in several dialects and languages and to learn new words along the way for continuous growth. This application of intelligence systems in the developments of the search engines may be useful in pursuing the application into the medical field and would actually be a time and cost saving methodology of the future. The results of this research may be more reflective in the developing nature of technology and the need of gadgets and other electronic mediums to register and translate voice commands and voice to speech facilities. The medical industry may need a tool that has substantial applications not just in transcriptions but in other forms of relevant functions as reports and patient files to have access available at any instant. Works Cited: Ayuso, A. J. & Soler, M. L. 1995. Speech Recognition and Coding: New Advances and Trends. Springer. Carter, J. H. 2001. Electronic Medical Records: A Guide for Clinicians and Administrators. American College of Physicians. Kiel, J. M. 2000. Information Technology for the Practicing Physician. Springer. Maheu, M. M. et al. 2004. The Mental Health Professional and the New Technologies: A Handbook for Practice Today. Lawrence Erlbaum. Ranchhod, E. 2002. Advances in Natural Language Processing: Third International Conference. Springer. Ribeiro, B. et al. 2005. Adaptive and Natural Computing Algorithms. Springer. Saito, S. & Shuzo, S. 1991. Speech Science and Technology. Ios Pr Inc. Sanfeliu, A. & Schulcloper, J. 2004. Progress in Pattern Recognition, Speech and Image Analysis: 8th Iberoamerican Congress on Pattern Recognition. Springer. Waibel, A. & Lee, K. 1990. Readings in Speech Recognition. Morgan Kaufmann. Read More