Repairing General-Purpose ASR Output to Improve Accuracy of Spoken Sentences in Specific Domains Using Artificial Development Approach / 4234
C. Anantaram, Sunil Kumar Kopparapu, Chirag Patel, Aditya Mittal

General-purpose speech engines are trained on large corpus. However, studies and experiments have shown that when such engines are used to recognize spoken sentences in specific domains they may not produce accurate ASR output. Further, the accent and the environmental conditions in which the speaker speaks a sentence may induce the speech engine to recognize certain words/ sets of words inaccurately. Thus, the speech engine's output may need to be repaired for a domain before any further natural language processing is carried out. We present an artificial development (Art-Dev) based mechanism for such a repair. Our approach considers an erroneous ASR output sentence as a biological cell and repairs it through evolution and development of the inaccurate genes in the cell (sentence) with respect to the genes in the domain. Once the genotypes are identified, we grow the genotypes into phenotypes to fill the missing gaps or erroneous words with appropriate domain concepts. We demonstrate our approach on the output of standard ASR engines such as Google Now and show how it improves the accuracy.

PDF