Wai Kit Lo

A unified approach in speech-to-speech translation: integrating features of speech recognition and machine translation

Based upon a statistically trained speech translation system, in this study, we try to combine distinctive features derived from the two modules: speech recognition and statistical machine translation, in a loglinear model. The translation hypotheses are then rescored and translation performance is improved. The standard translation evaluation metrics, including BLEU, NIST, multiple reference word error rate and its position independent counterpart, were optimized to solve the weights of the features in the log-linear model. The experimental results have shown significant improvement over the baseline IBM model 4 in all automatic translation evaluation metrics. The largest was for BLEU, by 7.9% absolute.

Generalized posterior probability for minimum error verification of recognized sentences

Generalized posterior probability (GPP) is investigated in this paper as a statistical confidence measure for verifying recognized sentences of a large vocabulary continuous speech recognition system (LVCSR). We optimize the GPP by training the exponential weights of the acoustic and language models and decision threshold to minimize total verification errors. Two utterance level confidence measures: generalized utterance posterior probability (GUPP) and product of generalized word posterior probabilities (GWPP) of component words in a string hypothesis are tested. When evaluated on the Chinese Basic Travel Expression Corpus (BTEC), 47.9% and 53.9% relative improvement of utterance confidence error rate (CER) have been obtained for the GUPP and product of GWPP confidence measures, respectively.

Generalized posterior probability for minimizing verification errors at subword, word and sentence levels

Generalized posterior probability, a statistical confidence measure, is tested in this study for verifying optimally the recognized units at the subword, word and sentence levels. We developed the generalized posterior probability by analyzing the exponential weights of the acoustic and language model scores to minimize the total verification errors at different unit levels. Experimental results have demonstrated the effectiveness of this generalized confidence measure for verifying Chinese LVCSR output. The Chinese Basic Travel Expression Corpus (BTEC) is used for evaluation and the relative improvement of confidence error rate (CER) over the baseline performance is 47.76% for sentences, 27.31% for words and 4.64% for subwords.