Effective Voice Enabled Interface for resource poor language (Gujarati)

Abstract

"Traditional approach consisting of ASR and IR is not much effective for short speech segment in the domain with higher word error rate[5].So IR system for spoken queries is one of the research areas in IR domain. Our aim is to correct query text for Voice-based resource poor language (Gujarati) and explore issues related to the creation of this system which combines state-of-the-art Speech recognition and information retrieval. In this thesis work, we have designed voice-enabled Sandhan App for Gujarati language. Transcribed queries may have errors which directly effects the performance of IR system. So to improve the performance, we first tried to correct the transcribed text using the concept of Soundex algorithm and Language modeling. To reduce access time, we developed Query word suggestion interface.In addition to previous methods, we developed Query reformulation box which consists addition, deletion and substitution options. This feature allows users to modify misrecognized words."