Hold Me Tight: Stable Encoder-Decoder Design for Speech Enhancement
This work addresses training challenges in speech enhancement models, offering a stable encoder-decoder design that is incremental but domain-specific.
The paper tackles the instability and training difficulty of 1-D convolutional filters in speech enhancement by combining auditory filterbank preprocessing, frame theory-based unsupervised objectives, and mixed compressed spectral norms, resulting in significant improvements in perceptual speech quality (PESQ).
Convolutional layers with 1-D filters are often used as frontend to encode audio signals. Unlike fixed time-frequency representations, they can adapt to the local characteristics of input data. However, 1-D filters on raw audio are hard to train and often suffer from instabilities. In this paper, we address these problems with hybrid solutions, i.e., combining theory-driven and data-driven approaches. First, we preprocess the audio signals via a auditory filterbank, guaranteeing good frequency localization for the learned encoder. Second, we use results from frame theory to define an unsupervised learning objective that encourages energy conservation and perfect reconstruction. Third, we adapt mixed compressed spectral norms as learning objectives to the encoder coefficients. Using these solutions in a low-complexity encoder-mask-decoder model significantly improves the perceptual evaluation of speech quality (PESQ) in speech enhancement.