Boosting Speech-to-Text software potential

Abstract

The article focuses on finding ways of boosting efficiency and accuracy of Speech-to-Text (STT)-powered input. The effort is triggered by the growing popularity of the software among professional translators, which is in line with the general trend of abandoning typing in favor of speech-to-text applications. Insisting that better effectiveness of such programs is contingent on their accuracy, the researchers analyze major factors, both linguistic and technical in nature, affecting the computer-assisted speech transcribing quality. This leads to an experiment, putting the hypothesis to a test. Based on numerical and performance data, errors and their breakdown into categories in an attempt to figure out their origins, it dwells on various approaches to dictation in a combination with several hardware options and configurations. These pave the way for recommendations on the improvement of STT performance based on the Dragon software. The authors arrive at a conclusion that it is possible to boost the STT accuracy up to 99 percent by adjusting the program profile to accommodate phonetic features of the speaker with due consideration of his accent, adding to the dictionary the most complex and rare vocabulary beforehand, and fine-tuning input hardware. Other noteworthy results include ways to overcome the most complex transcribing challenges, i.e. proper names, placenames, abbreviations, etc.