Research On The Matching Method Of Large Vocabulary Speech

Due to the characteristics of the voice and the needs of modern society, researches on the recognition of speech have a great deal of academic and practical values. But the existing large vocabulary speech is far from the hopes of people. The problem is mainly in matching method, and solving the matching method of large vocabulary speech recognition is particularly important.The purpose of this thesis is to solve the slow speed of large vocabulary recognition by reducing the number of matches, and by building a database for speeding up the search, improves the speed of such a large vocabulary recognition speed. This thesis mainly focues on dividing vowel, recognition vowel, judgment tone, and build databases.(1) On the division of vowel and consonant By calculating maximum energy to experiment on different syllables pronunced sound, the same pronounced syllables and only composed of vowel and each syllable in different classes with different combinations of tones.Obtaining the maximum distance of energy is effective for division of three categories; different vowels impact on the division and it can be overcome by changing speech rate; different tones have no impact on the division.(2) On the recognition of vowel and consonant Determine the mel cepstral as the characteristic parameter by undersdanding mel cepstral, matching calculate using the correlation function, the feature of each consonant is extracted and do correlated operations with respective template. The maximum correlation coefficient is the recognition result.(3) Judgeing tone Understanding the fundamental characteristics, decided to extract the fundamental frequency of vowel to simulate the curve, then set rules and determine a few tones according to the laws of the curve.(4) On the establishment of a database Establish libraries and font characteristics, the identified information by the signature consonants, vowels and tones to address as a bridge leading to the store, fonts stored in accordance with the grouping, experiments show that the construction method of Chinese characters speed up the search speed.