Cross-Domain Diffusion Based Speech Enhancement for Very Noisy Speech.
Heming WangDeLiang WangPublished in: ICASSP (2023)
Keyphrases
- cross domain
- speech enhancement
- noisy speech
- noise reduction
- speech recognition
- noisy environments
- signal to noise ratio
- single channel
- speech signal
- hidden markov models
- background noise
- anisotropic diffusion
- linear prediction
- vocal tract
- transfer learning
- wiener filter
- knowledge transfer
- smoothing algorithm
- multi channel
- edge detection
- edge preserving
- multiscale
- automatic speech recognition
- image restoration
- machine learning
- random fields
- non stationary
- prior knowledge