Mid-attribute speaker generation using optimal-transport-based interpolation of Gaussian mixture models.
Aya WatanabeShinnosuke TakamichiYuki SaitoDetai XinHiroshi SaruwatariPublished in: CoRR (2022)
Keyphrases
- gaussian mixture model
- speaker recognition
- speaker identification
- mixture model
- em algorithm
- maximum likelihood
- speaker verification
- expectation maximization
- feature vectors
- probability density function
- feature space
- covariance matrices
- density estimation
- gaussian mixture
- vector quantization
- variational bayes
- mel frequency cepstral coefficients
- data mining
- finite mixtures
- closed form
- prior knowledge