Potentials for ASR based on Multiple Acoustic Models and Model Selection using Standard Speech Features.
Thomas WinklerDaniel SteinRolf BardeliDaniel SchneiderJoachim KöhlerPublished in: ITG Conference on Speech Communication (2012)
Keyphrases
- model selection
- speech recognition
- automatic speech recognition
- cross validation
- acoustic models
- speech signal
- broadcast news
- model selection criteria
- speech recognition systems
- feature extraction
- selection criterion
- marginal likelihood
- feature selection
- mixture model
- active learning
- prior knowledge
- speech recognizer
- machine learning
- noisy environments
- data sets
- speaker identification
- high dimensional
- dialogue system
- gaussian process
- pattern recognition
- classification accuracy
- probabilistic model
- feature vectors