ACL RD-TEC 1.0 Summarization of H93-1020
Paper Title:
ON THE USE OF TIED-MIXTURE DISTRIBUTIONS
ON THE USE OF TIED-MIXTURE DISTRIBUTIONS
Authors: Owen Kimball and Mari Ostendorf
Primarily assigned technology terms:
- acoustic modeling
- algorithm
- baum-welch algorithm
- bootstrapping
- clustering
- computing
- context modeling
- continuous speech recognition
- database
- em algorithm
- expectation-maximization
- hill climbing
- hillclimbing
- hmm system
- hmms
- iterative expectation-maximization
- iterative hillclimbing
- iterative segmentation\/estimation
- k-means
- k-means clustering
- load balancing
- mapping function
- marginal training
- maximum likelihood
- mixture modeling
- mixture training
- modeling
- n-best rescoring
- normalization
- parallel training
- parameter estimation
- parameter re-estimation
- parameterization
- processing
- processor
- re-estimation
- recognition
- recognition systems
- rescoring
- resource management
- segmentation
- speaker-independent recognition
- speech recognition
- speech recognition \
- ssm training
- synchronization
- terminology
- tied-mixture context modeling
- time warping
- training algorithm
- vector quantization
- viterbi
- viterbi training
- weight estimation
- weighting
Other assigned terms:
- alphabet
- approach
- bias
- byblos system
- case
- cluster
- conditional independence
- context models
- continuous speech
- convergence
- correlation
- data set
- dimensionality
- distribution
- error rate
- estimation
- feature
- feature sets
- feature space
- feature vectors
- formalism
- frame
- frequency correlation
- gaussian distribution
- gaussian mixture
- hypotheses
- implementation
- independence assumption
- likelihood
- likelihood parameter
- linear time
- mapping
- maps
- markov chain
- method
- mixture models
- model parameters
- model size
- parallelism
- probabilities
- probability
- probability density
- procedure
- process
- sentence
- sentences
- statistical approach
- statistics
- stochastic segment model
- syllables
- technique
- term
- test data
- test set
- training
- training corpus
- training data
- triphone
- utterance
- word
- word error rate
- words
