Continuous Speech Recognition and Verification based on a Combination Score

摘要

In this paper we present a speech recognition and verification method based on the integration of likelihood and likelihood ratio. Speech recognition and verification is unified in one-phase framework. A modified agglomerative hierarchical clustering algorithm is adopted to train the alternative model used in speech verification. In the process of decoding likelihood ratio is combined with likelihood to get the combination score for searching the final results. Our experimental results showed that false-alarm rate get decreased a lot with only slight loss in accuracy rate.