SPEECH RECOGNITION SYSTEM FOR WRITING DENTIST MEDICAL RECORDS

• Main Author: Yora Islami, Dinda
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/50846
• Institution: Institut Teknologi Bandung

Time constraints prevent dentists from making complete medical records. In overcoming these limitations, a speech recognition system in dentistry domain was developed to write patient medical records. It requires a dental speech corpus, a dental text corpus and proper techniques in develop acoustic models for the speech recognition system. In this final project discusses the development of a dental speech corpus, a dental text corpus, development of an acoustic model and a language model to be a solution to these problems. The text corpus is obtained from medical record scenarios developed by dentists. The speech corpus is obtained by recording sound to 100 people where each person speaks the predetermined text. In this final project, in building an acoustic model, a comparison is made between the GMM-HMM technique and the CNN-HMM technique to select the best technique. The text corpus and speech corpus that have been collected are then used as input in developing the speech recognition system. The construction of acoustic models uses the GMM-HMM and CNN-HMM techniques. System testing is done by calculating the word error rate (WER) value, the lower the WER value, the better the system. The acoustic model that has been developed is evaluated and compared with the WER value. The WER value in the CNN-HMM technique resulted 8.61% reduction compared to the GMMHMM technique. The solution of this final project uses the CNN-HMM technique because it has better performance than GMM-HMM.