Non-negative temporal decomposition of speech parameters

Abstract

We present a non-negative temporal decomposition method for line spectrum pair and articulatory parameters. Based on the multiplicative update rules derived from a non-negative matrix factorization algorithm, these parameters decompose into a set of temporally overlapped event functions that are restricted to the range [0, 1] and corresponding event vectors. With the proposed method, the RMS error of the measured and estimated articulatory parameters is 0.16 mm and the spectral distance of the measured and estimated line spectrum pair parameters is 1.97 dB. These results also show that these estimation errors of proposed method are significantly smaller than those of the conventional method. This technique will be useful for many applications, such as speech coding and speech modification.

Advanced Research in Speech, Language, and Hearing for Communication of the Future

“…Here as well, the combination of multiple technologies will give rise to new technologies that have a competitive advantage, and the germination of these technologies has already started. "Speaking Rhythm Extraction and Control by Nonnegative Temporal Decomposition" [4] is an achievement born of research into the mechanism of human speaking. The human voice is generated as a sound originating in the vibration of speech organs, and we are working on fascinating developments by combining that process with information-science technologies related to speech processing.…”