Recognition tone and syllable in combination for Lao continuous speech

Abstract

This thesis proposed a robust method on tonal syllable recognition for continuous Lao speech recognition, by applying the specific characteristic of Lao language system as the conditional classification. The Lao language was studied in both acoustical and grammatical of Lao spoken. From the acoustical point of view, a syllable consists of initial consonant, vowel, final consonant and a tone. The final consonant is strongly influenced by the vowel duration. In addition, the part of vowel and final consonant is considered as voiced portion. Usually, only the voiced portion can carry tone information. Since, the initial consonant is not affected by the duration of the vowel. From the grammatical point of view, the tone of a syllable will be known, when the initial consonant, vowel and final consonant are know. Furthermore, the meaning of some syllables can change when different tones are applied. Therefore, this research investigated to increase the accuracy and recognition speed of the Lao speech recognition by applying the condition of Lao tonal grammar. To implement a tonal syllable recognition system for Lao language based on continuous speech, various conventional speech units used in speech recognition systems have been investigated, in order to find out the optimal speech model for Lao speech recognition. In experiments, the existing techniques of tonal syllable recognition have also been experimented to evaluate the effectiveness and compare with the proposed method. As a result, experiment results shown that, the proposed system can achieve higher recognition rate at 66.85% and 8.92%, for speaker-independent and speaker-dependent, respectively. Moreover, the proposed system can recognizes with faster speed than that of baseline system, around 25%.