The high cost of data entry is a critical issue that has challenged the evolution of computerized patient record systems. Development of dynamic language models is proposed for significantly improving the cost performance of medical transcription systems. The innovative use of speech recognition, computer telephony integration, and the Internet is proposed for the management and transcription of physician dictation. Preliminary market research and analysis have already demonstrated the potential for significant cost savings to healthcare organizations. Our goal is to apply multiple dynamic language models to both improve accuracy and the robustness of the system. We propose to explore different methodologies for creating categories for sets of words to reduce grammar size, research means to reduce Word Error Rate (WER), determine the best strategies for improving recognition hypotheses confidence, experiment with different language model types, discover appropriate initial weightings for the various models, and calculate the optimal update frequency for the language models. Our unique approach integrates seamlessly into a physician's workflow and does not require the alteration of physician work patterns. We expect this development will result in a functional prototype of a transcription system that is commercially viable and designed to reduce the costs associated with medical transcriptions. eScription has obtained two paying pilot customers who we are working closely with us to develop this prototype. The two hospitals have/will provide eScription with textual data, audio data, and medical transcriptionist who will test the final system. We are submitting this grant request to partially cover the cost of constructing and testing of language models. We are independently funding a research effort concerning audio models that includes analyzing the impact of a speaker's accent.
eScription focuses on alleviating significant healthcare cost pressures associated with transcription of medical dictation. We apply new technologies such as speech recognition, compute telephony and Internet communications which are not commonly used for medical transcription, We will directly sell our software products and services to Integrated Delivery Networks (IDNs) and to Transcription Services Companies,
