No abstract

The frame rate of the observation sequence in distributed speech recognition applications may be reduced to suit a resource-limited front-end device. In order to use models trained using full-frame-rate data in the recognition of reduced frame-rate (RFR) data, we propose a method for adapting the transition probabilities of hidden Markov models (HMMs) to match the frame rate of the observation. Experiments on the recognition of clean and noisy connected digits are conducted to evaluate the proposed method. Experimental results show that the proposed method can effectively compensate for the frame-rate mismatch between the training and the test data. Using our adapted model to recognize the RFR speech data, one can significantly reduce the computation time and achieve the same level of accuracy as that of a method, which restores the frame rate using data interpolation.

No abstract

No abstract

This paper presents a method of feature extraction for the automatic recognition of voiceless unaspirated stop consonants in Mandarin speech. The features are derived from the spectrographic acoustic patterns of syllable-initial voiceless unaspirated stops /p,t,k/, which include the burst spectrum, the formant transition, and the voice onset time. A normalization process for the second and the third formants at the voice onset is proposed. Based on these derived features, Bayes classifiers and a layered neural net are applied to classify the places of articulation of these stop consonants. The experiments show that the derived features are robust and efficient for speaker-independent speech recognition, and the neural net is a preferable choice in the classification of these stops in multiple contexts.