Skip to Main Content
In this paper, a syllable-based Chinese speech recognition system is described which consists of components for carrying out three main tasks: syllable segmentation, acoustic analysis and training. Connected speech is first divided into syllabic units, and an efficient short-time FFT algorithm with two different analysis frame sizes is then used to analyse the units in order to find the dynamic spectrum of syllables. Finally an important feature, sound-stimulation, is extracted from syllable dynamic spectrum and assembled in one feature matrix for each syllable. The proposed method based on extraction of sound-stimulation emphasizes phoneme-phoneme transition properties within syllable, and has proved effective and feasible in experiment for Chinese speech recognition.