Research on acoustic parameters of singing voice for singing speech intelligence

Abstract

In artificial intelligence research, prosodic and pitch features are the focus of singing intelligent speech synthesis. Pitch features are usually studied by extracting fundamental frequencies. In the process of folk song singing, many special vocal skills are used, mainly related to the mode of vocal fold vibration. Using singing voice signals and parameters to study singing pitch and vocal types can better obtain the pitch and prosodic characteristics of singing. This experiment mainly takes Ewenki folk songs as the research object, analyzes singing voice signals and parameters, so as to provide reference for singing voice synthesis. The Ewenki are one of China’s ethnic groups with a small population. The Ewenki folk songs are sung in the Ewenki language, which retains many features of ancient folk songs and singing pronunciation characteristics, especially in the vocal vocalization of the singing voice. The article takes the typical traditional folk song of Ewenki, "Molie Girl", as the object of study. By collecting the singing voice and voice signals of a professional and a non-professional singer, extracted the parameters of fundamental frequency, open quotient and speed quotient. After statistical analysis of the parameters, the overall voice characteristics of the professional and non-professional singers during the singing process and the differences in the voice parameters of the trailing part of the voice were obtained. The analysis revealed both professional and non-professional singers were able to express the artistic characteristics of traditional folk songs, with different vocal styles and singing techniques. The experimental voice acoustic parameters and results can be used as a reference for the synthesis of traditional folk songs and as a method for the teaching of traditional folk songs.