Journal of Intelligent Procedures in Electrical Technology (Oct 2013)
Syllable Segmentation of Farsi Continuous Speech using Wavelet Coefficients Thresholding and Fuzzy Smoothing of Energy Contour
Abstract
Syllable, as a sub-word unit, nowadays plays an active role in the field of speech processing and recognition research according to its robust relation to human speech production and cognition. Automatic syllable boundaries detection is an important step forward in the areas of speech prosody, natural speech synthesis and speech recognition. In this paper, a novel method in automatic syllabification of Farsi continuous speech based on acoustic structure is proposed. Our previous studies, showed the proficiency of energy contour fuzzy smoothing method, compared with other prominent works in this area. This paper suggests that the conventional methodology-used in speech enhancement based on wavelet coefficient thresholding would improve syllable segmentation by decreasing insertion error. This process declines the energy in high energy consonants which are responsible for extra peaks in short term energy contour. Experimental results showed that utilizing proposed method along with fuzzy smoothing would diminish insertion error about 8% with no reasonable effect on other efficiency criteria. More than 94% of syllables are automatically segmented using presented technique with less than 50ms error.
