A quantitative method for performance analysis of an isolated word ASR system.
Michele GubianLuigi ArnoneSergio BrofferioPublished in: EUSIPCO (2005)
Keyphrases
- cost function
- error rate
- high precision
- significant improvement
- main contribution
- theoretical analysis
- high accuracy
- evaluation method
- fully automatic
- synthetic data
- detection method
- non stationary
- model selection
- classification accuracy
- experimental evaluation
- similarity measure
- neural network
- semi supervised
- data sets
- feature set
- computational cost
- detection algorithm
- clustering method
- hidden markov models
- k means
- training set
- pattern recognition
- mathematical model
- objective function