[Computerized speech recognition-based endoscopic findings].

Abstract

Discrete, Hidden Markov model based speech recognition and phoneme based speech synthesis techniques were applied for gastroscopy reporting and machine control. The authors developed a special program for grammatical analysis of the sentences. Altogether 100 patient findings were grammatically analysed. The sentences were grouped according to the topographical order of the investigation: oesophagus, cardia, fundus, corpus, antrum, pylorus, bulbus, postbulbar section, and the pathological findings: erosion, ulceration, malignancy. Speech samples from 3 deep voiced male investigators were collected. The recognition rate was above 95%. A simulation program was also developed for dictation and controlling of the different equipment (monitor, printer, video, endoscope) in the gastroscopy laboratory by speech recognition. Speech synthesis was applied for the evaluation of understanding. This module artificially synthesizes the answer of the system giving backup for the understood information. With additional developments the discrete word speech 'recognition' achieved the level of routine application in medical reporting. However, ready-to-use developments need the joint activity of speech technology and endoscopy industry with end-user teams.