Combining Adversarial Training and Disentangled Speech Representation for Robust Zero-Resource Subword Modeling.
Siyuan FengTan LeeZhiyuan PengPublished in: CoRR (2019)
Keyphrases
- speech recognition
- broadcast news
- hearing impaired
- model construction
- speech signal
- audio visual
- computationally efficient
- training process
- resource allocation
- partial occlusion
- training samples
- image representation
- text classification
- training set
- n gram
- test set
- training data
- web resources
- training algorithm
- online learning
- dynamic bayesian networks
- noisy environments
- supervised learning
- spoken document retrieval