Non-Negative Temporal Decomposition of Speech Parameters by Multiplicative Update Rules

Dysarthric Speech Modification Using Parallel Utterance Based on Non-negative Temporal Decomposition

“…(2) is minimized by iteratively updating (3) -(5), which is shown at the top of the next page. These updating rules are derived in [11]. In [11], line spectral pair (LSP) is used as a spectral feature; however, we use a magnitude spectrum to estimate the event basis and activity more precisely.…”

“…These updating rules are derived in [11]. In [11], line spectral pair (LSP) is used as a spectral feature; however, we use a magnitude spectrum to estimate the event basis and activity more precisely. Moreover in [11], each event basis corresponds to a single phoneme.…”

“…In [11], line spectral pair (LSP) is used as a spectral feature; however, we use a magnitude spectrum to estimate the event basis and activity more precisely. Moreover in [11], each event basis corresponds to a single phoneme. In order to estimate the event basis and activity more precisely, 3 event bases are extracted from a single phone.…”

“…This paper proposes a dysarthric speech modification system using parallel utterances in order to improve the intelligibility of dysarthric utterances. Non-negative Temporal Decomposition (NTD) [11] has been proposed in the field of speech coding and it is applied to the rhythm conversion of non-native English. We applied NTD to dysarthric speech.…”

Dirichlet Latent Variable Model: A Dynamic Model Based on Dirichlet Prior for Audio Processing

“…Efforts have been put towards learning dynamic models by imposing temporal constraints on the bases as well as on their coefficients. The dynamic models include sparse and dynamic variant of LVM/NMF [8], [9], [10], [11], convolutive NMF [12], [13], [14] and non-negative hidden Markov model (NHMM) [15].…”