The Acoustic Performance Of Speech Rhythmic Phrase Boundaries And The Study Of Speech Pause Recognition

The rhythm structure information is very important in speech synthesis and speech recognition.There are two important indicators which are the degree of natural and intelligibility respectively in speech synthesis.At present,the current intelligibility has been up to standard while the degree of nature is not enough.The reason is that the computer can not accurately identify the rhythm of the sentence information.In speech recognition,the computer is required to be able to accurately identify the speech in the statement pause and the sentence is automatically divided in order to realize the natural exchange of man-machine,make computer understand the language of the person,and accurately distinguish the meaning of the statement or the speaker thus the speaker statement can convert into the machine language and operate accurately in accordance with the meaning of the statement.Therefore,from the perspective of acoustics,the acoustic characteristic parameters are extracted directly from the speech in this paper.The acoustic characteristic parameters at the boundary of the prosodic phrase are analyzed.Based on the acoustic characteristics,the model is constructed to realize the speech rhythm pause.The main work of this paper has the following three parts:(1)Text processing and audio feature extraction.The word corpus is worded,and the rhythm boundary which can not exist is removed by the word segmentation.The word after the word segmentation is converted into its corresponding phonetic string based on the Chinese character-phonetic dictionary.Based on the speech corpus,we can obtain the acoustic parameters such as short-time energy,short-term amplitude,short-time zero-crossing rate,fundamental frequency,center of mass,spectral entropy,information entropy,phoneme vowel and so on and extract the relevant waveform curve or data of acoustic parameters.(2)Analysis of acoustic performance at the boundary of phonetic rhythm.Based on the related waveforms or data of the acoustic characteristics,the acoustic performance at the phonetic pause is preliminary analyzed.Then,the acoustic performance of the Chinese rhythm boundary is further analyzed based on the combination or transformation of the acoustic characteristic parameters..(3)Speech Pause Recognition Based on Acoustic Characteristics.Firstly,based on the acoustic performance of the prosodic boundary,the candidate acoustic feature set is constructed.Then the characteristic template is constructed by selecting the appropriate acoustic feature.Finally,the support vector machine model is used to realize the automatic recognition of the prosodic pause,and the experimental results are analyzed.