Unsupervised acoustic unit discovery for speech synthesis using discrete latent-variable neural networks.
Ryan EloffAndré NortjeBenjamin van NiekerkAvashna GovenderLeanne NortjeArnu PretoriusElan Van BiljonEwald van der WesthuizenLisa van StadenHerman KamperPublished in: CoRR (2019)
Keyphrases
- speech synthesis
- latent variables
- prosodic features
- neural network
- speech recognition
- global exponential stability
- text to speech
- probabilistic model
- vocal tract
- random variables
- latent variable models
- unsupervised learning
- topic models
- pattern recognition
- hidden variables
- semi supervised
- back propagation
- prior knowledge
- hierarchical model
- topic modeling
- knowledge discovery
- observational data
- gaussian process
- real valued
- hidden markov models
- observed variables
- weakly supervised
- decision trees
- supervised learning
- probabilistic principal component analysis
- speaker verification
- hopfield neural network
- probabilistic latent semantic analysis
- causal relationships
- image processing