Acoustic feature analysis and conversion of age speech

Statistical property of acoustic features of age speech spoken by different age speakers and genders is described in this paper. Spectrum features including formant frequencies (F1, F2) and prosodic features including fundamental frequency (F0) and speaking rate are analyzed based on age speech corpus SUDA_NL which contains totally 256-minute samples of four age speaker groups as children, youth, middle and old age respectively. Furthermore, an age speech conversion system using universal background model (UBM) is also proposed. The analysis results show the acoustic feature parameters have specific trends with age increasing and the UBM-based age speech conversion is effective according to subjective and objective evaluations.