Bayesian Parametric and Architectural Domain Adaptation of LF-MMI Trained TDNNs for Elderly and Dysarthric Speech Recognition.
Jiajun DengFabian Ritter GutierrezShoukang HuMengzhe GengXurong XieZi YeShansong LiuJianwei YuXunying LiuHelen MengPublished in: Interspeech (2021)
Keyphrases
- speech recognition
- domain adaptation
- automatic speech recognition
- covariate shift
- isolated word
- discriminative training
- hidden markov models
- speech signal
- language model
- cross domain
- pattern recognition
- semi supervised
- multiple sources
- sentiment classification
- speech recognition systems
- transfer learning
- test data
- labeled data
- training data
- unlabeled data
- maximum likelihood
- training set
- bayesian networks
- training process
- neural network
- semi supervised learning
- probabilistic model
- document classification
- face recognition
- artificial intelligence
- text classification