15th International Congress of Phonetic Sciences (ICPhS-15) Barcelona, Spain August 3-9, 2003

Phonetic-Acoustical Problems of Personal Voice Cloning by TTS

Boris M. Lobanov (1), Helena B. Karnevskaya (2)

(1) National Academy of Sciences, Belarus
(2) Minsk Linguistic State University, Belarus

The report describes a recent development of a TTS-system for Russian and English based on allophonic natural speech signal elements (about 1500 in all) with the maximal possible imitation of individual male and female voices. In distinction to the biological task of cloning, the target is not a copy of the human being as a whole but of only one of its functions, particularly, that of reading aloud an orthographically unrestricted text preserving thereby the individual acoustic characteristics of a speaker's voice, as well as his/her phonetic (segmental and prosodic) peculiarities. A successful solution of the task outlined above presupposes that the following two requirements should be unequivocally satisfied:

• The fullest possible use of a complex of acoustic characteristics carrying information about the individual voice and pronunciation properties of the speaker being imitated.
• The minimal possible distortions of the elements of concatenation at all stages of their 'production', as well as the maximal possible accuracy of prosodic modifications in the process of speech synthesis.

Full Paper

Bibliographic reference.  Lobanov, Boris M. / Karnevskaya, Helena B. (2003): "Phonetic-acoustical problems of personal voice cloning by TTS", In ICPhS-15, 2565-2568.