J. Data and Text Mining, Natural Language Processing

摘要

We compare the performance of the computerized speech recognition system to convert voice medical radiological reports into text with the conventional hand transcription. We used the "Dragon Naturally Speaking Preferred Spanish version number 10" program. Automatic Speech recognition decreased the transcription time and reduced substantially the transcription costs. Speech recognition by computer is practical in radiological reports implementation. In our study, we found that hand writing transcription of digitized files of short Spanish radiological reports has an error of 5% compared to the gold standard. This grows to 13% when using the automatic speech recognition system, only 8% difference. Automatic speech recognition programs are useful for text transcription of short radiological reports in Spanish. We cannot say the same for other languages or clinical applications and, in each case, it would be advisable to test its efficiency.