Presentation | 6th Internet World Congress for Biomedical Sciences |
Marcial García Rojo(1), Ana Morillo Castro(2)
(1)Servicio de Anatomía Patológica. Complejo Hospitalario de Ciudad Real - Ciudad Real. Spain
(2)Centro de Salud III - Ciudad Real. Spain
Contact address: |
Marcial García Rojo Servicio de Anatomía Patológica Complejo Hospitalario de Ciudad Real Avda´ Pío XII s/n Ciudad Real Ciudad Real 13002 Spain marcial@cim.es |
The main obstacles in speech recognition are linguistic variability, factors related to user (Pronunciation, Inflexion…), channel variability and phonemes context. To overcome these difficulties, we describe the basic acoustic and language models to understand the basics of speech recognition. We have studied and tested some of the products available in speech recognition in English and Spanish (Philips FreeSpeech, IBM ViaVoice98, L&H Voice Xpress, Dragon Naturally Speaking. We found important differences in the user interface, and recognition abilities in each of the programs evaluated. In every case, there is a small percentage of misinterpretation that must be corrected by the user.
In order to avoid these small but uncomfortable mistakes in the recognition process we recommend the use of macros.
We conclude that these systems can already be used in clinical areas, and in the next few years speech recognition software will be a usual part of our daily work, and it will be helping us not only in a more effective clerical work, but also in medical activities like the activation of instruments during surgery or the integration in wireless systems at patient´s bedside, such as small portable computers used in the recollection of patients´ data and direct transmission to central server.