A Cross-Entropy-Guided Measure (CEGM) for Assessing Speech Recognition Performance and Optimizing DNN-Based Speech Enhancement.
Li ChaiJun DuQing-Feng LiuChin-Hui LeePublished in: IEEE ACM Trans. Audio Speech Lang. Process. (2021)
Keyphrases
- speech recognition
- cross entropy
- speech enhancement
- noisy environments
- speech signal
- noisy speech
- language model
- spectral subtraction
- hidden markov models
- speech synthesis
- language modeling
- automatic speech recognition
- vocal tract
- log likelihood
- speaker identification
- scoring function
- maximum likelihood
- noise reduction
- pattern recognition
- linear prediction
- additive noise
- background noise
- signal to noise ratio
- evaluation metrics
- multiscale
- training process
- evaluation measures