Improving the Utility of Speech Recognition Through Error Detection

Abstract

Despite the potential to dominate radiology reporting, current speech recognition technology is thus far a weak and inconsistent alternative to traditional human transcription. This is attributable to poor accuracy rates, in spite of vendor claims, and the wasted resources that go into correcting erroneous reports. A solution to this problem is post-speech-recognition error detection that will assist the radiologist in proofreading more efficiently. In this paper, we present a statistical method for error detection that can be applied after transcription. The results are encouraging, showing an error detection rate as high as 96% in some cases.