Direct acoustic feature using iterative EM algorithm and spectral energy for classifying suicidal speech.
T. YingthawornsukH. Kaymaz KeskinpalaD. Mitchell WilkesRichard G. ShiaviRonald M. SalomonPublished in: INTERSPEECH (2007)
Keyphrases
- em algorithm
- expectation maximization
- mixture model
- maximum likelihood
- speech sounds
- parameter estimation
- gaussian mixture model
- expectation maximisation
- generative model
- likelihood function
- maximum a posteriori
- incomplete data
- log likelihood
- gaussian mixture
- spectral features
- model based clustering
- density estimation
- maximum likelihood estimation
- mixture modeling
- feature vectors
- speech recognition
- probability density function
- feature set
- acoustic features
- finite mixture model
- hidden variables
- image features
- likelihood maximization
- penalized likelihood
- matrix factorisation
- dimensionality reduction
- computer vision