Modelling Feature Vectors
Modelling Feature Vectors
- Stationary regions
- Voiced regions, especially vowels in isolated word speech
- Silence
- Transient regions
- Plosives
- Most sounds in continuous speech
- What should we model?
- Stationary? (e.g GMM)
- Quasi-stationary? (e.g. HMM)
- Non-stationary? (e.g. SSM, HMM variants)
- State-of-the-art:
- Hidden Markov Model (HMM)
- Future work:
- HMM extensions for non-stationary modelling
- Trajectory / Segment based models