ACL RD-TEC 1.0 Summarization of H94-1079
Paper Title:
THE LINCOLN LARGE-VOCABULARY STACK-DECODER BASED HMM CSR
THE LINCOLN LARGE-VOCABULARY STACK-DECODER BASED HMM CSR
Primarily assigned technology terms:
- algorithm
- bayesian smoothing
- beam pruning
- binary split
- binary splitting
- bootstrapping
- bootstrapping procedure
- caching
- clustering
- continuous-speech recognition
- data reduction
- database
- decoder
- decoding
- depth-first search
- df search
- error reduction
- estimation algorithm
- fast match
- forward-backward algorithm
- grouping
- hmm csr
- language modeling
- linear interpolation
- linking
- maximum-likelihood
- ml estimation
- modeling
- path search
- pruning
- recognition
- recognition search
- recognition system
- recognition-time adaptation
- recognizer
- search
- search strategy \
- search technique
- searching
- simulated annealing
- smoothing
- smoothing weight estimation
- speaker adaptation
- speaker-independent training
- splitting
- stack decoder
- training procedure
- transcription
- tree clustering
- tree search
- triphone smoothing
- viterbi
- viterbi decoder
- viterbi training
- weight estimation
Other assigned terms:
- acoustic models
- approach
- array
- background model
- beam
- bias
- buffer
- cache
- case
- cluster
- continuous speech
- continuous-speech
- cpu time
- data structures
- decision rule
- dictionary
- distribution
- duration
- error rate
- estimation
- fact
- implementation
- interpolation
- language model
- leaf
- likelihood
- log-likelihood
- method
- model parameters
- n-gram
- noise
- orthographic transcription
- partial theory
- probabilities
- probability
- procedure
- process
- recognition accuracy
- search problem
- search strategy
- sentence
- sentences
- signal
- silence
- stochastic language model
- system design
- technique
- term
- terms
- text
- theories
- theory
- tm system
- training
- training data
- tree
- triphone
- user
- utterance
- vocabulary
- word
- word error rate
- words
