A noise-robust ASR back-end technique based on weighted viterbi recognition.
Xiaodong CuiAlexis BernardAbeer AlwanPublished in: INTERSPEECH (2003)
Keyphrases
- back end
- noisy environments
- automatic speech recognition
- noise reduction
- building blocks
- data management
- hidden markov models
- recognition rate
- partial occlusion
- user friendly
- robust recognition
- image noise
- recognition process
- pose invariant face recognition
- data types
- speech recognition
- object recognition
- feature extraction
- spoken words
- version control
- salt pepper
- imaging artifacts
- varying illumination conditions
- data sets
- registration errors
- noise level
- information retrieval
- databases
- geometric distortions
- estimation error
- anti noise
- signal to noise ratio
- recognition accuracy
- illumination invariance