Study On Preserved Timbre Of Singing Harmonic Generation Method

In recent years,stimulated by the rapid development of mobile internet applications and various large-scale singing reality shows?,singing APPs are flourishing.At the same time,it also brings new challenges to singing voice synthesis and sound eff-ect processing system.The traditional sound effects have usually focused on enriching the space and the sense of hearing for the singer's voice.In the application scene of singing songs,the vocal accompaniment is most commonly used for enriching performance,which is due to some users were lacking of singing ability.However,most of singing APPs remain in using traditional audio effect processing method,and the singing harmonic has generated without according the user's singing ability.This paper generates the harmonic of singing as the main research objectives,takes the singer's unaccompanied singing audio signal as research object,selects the singer's vocal range and timbre as its characterization,and then builds the harmonic generation model.and ultimately applies to personalized audio effect processing.The system generates suitable harmonic by the representations of user's singing ability.Specifically,this paper includes the following three parts:Firstly,this paper presents a method of sentence boundary detection for singing voice.This method which is according to the human auditory perception properties for fine-grained endpoint detection,it is based on the use of singing has long pauses between sentences.This method analyzes the sentence boundary using K-means clustering algorithm.Experiments show that this method achieves a good performance to detect the boundary of sentence,the accuracy is more than 70%,and be able to adapt different musical rhythm.Secondly,this paper tries to analyze the representations of the voice timbre;we present a pitch shift method with preserved timbre for singing voice.Traditional methods have some problems,such as affecting the timbre,distortion,ignoring the transition and so on.We propose an improved algorithm which is based on linear prediction pitch shift,and optimizing the process for singing voice.Through the objective indexes analyzing and subjective evaluating experiments,the results show that the method we proposed for changing the pitch of singing voice can shift the pitch with preserved timbre,and the singing voice quality is greatly improved.Finally,this paper explores the harmonic of singing,and proposes a method of harmonic generation based on rhythm of singing.This method detects the tempo of singing based on spectral flux and self-adaptively adjusts the delaytime of the harmonic part according to the tempo of singing.Experiments show that this method can effectively extract the tempo of singing and reduce the time complexity,and the generated harmonic can enrich the musical expressions of the soloist.