iDeepMMSE: An improved deep learning approach to MMSE speech and noise power spectrum estimation for speech enhancement.
Minseung KimHyungchan SongSein CheongJong Won ShinPublished in: INTERSPEECH (2022)
Keyphrases
- speech enhancement
- power spectrum
- deep learning
- noisy environments
- noise reduction
- frequency domain
- single channel
- signal to noise ratio
- speech signal
- natural images
- vocal tract
- linear prediction
- noisy speech
- unsupervised learning
- background noise
- wiener filter
- speech recognition
- machine learning
- spatial domain
- additive noise
- denoising
- multi channel
- automatic speech recognition
- input image
- image denoising
- edge detection
- principal component analysis
- object recognition