Individuality-preserving Voice Conversion for Articulation Disorders Using Dictionary Selective Non-negative Matrix Factorization

We present in this paper a voice conver-sion (VC) method for a person with an ar-ticulation disorder resulting from athetoid cerebral palsy. The movements of such speakers are limited by their athetoid symptoms, and their consonants are of-ten unstable or unclear, which makes it difficult for them to communicate. In this paper, exemplar-based spectral con-version using Non-negative Matrix Factor-ization (NMF) is applied to a voice with an articulation disorder. In order to pre-serve the speaker’s individuality, we use a combined dictionary that was constructed from the source speaker’s vowels and tar-get speaker’s consonants. However, this exemplar-based approach needs to hold all the training exemplars (frames), and it may cause mismatching of phonemes between input signals and selected ex-emplars. In this paper, in order to re-duce the mismatching of phoneme align-ment, we propose a phoneme-categorized sub-dictionary and a dictionary selec-tion method using NMF. The effective-ness of this method was confirmed by comparing its effectiveness with that of a conventional Gaussian Mixture Model (GMM)-based and conventional NMF-based method.