Speech Denoising in the Waveform Domain with Self-Attention
This work addresses speech denoising for audio processing applications, representing an incremental improvement with a novel method for a known bottleneck.
The authors tackled speech denoising in the waveform domain by proposing CleanUNet, a causal model that combines encoder-decoder architecture with self-attention blocks and multi-resolution losses, achieving state-of-the-art results in objective and subjective metrics.
In this work, we present CleanUNet, a causal speech denoising model on the raw waveform. The proposed model is based on an encoder-decoder architecture combined with several self-attention blocks to refine its bottleneck representations, which is crucial to obtain good results. The model is optimized through a set of losses defined over both waveform and multi-resolution spectrograms. The proposed method outperforms the state-of-the-art models in terms of denoised speech quality from various objective and subjective evaluation metrics. We release our code and models at https://github.com/nvidia/cleanunet.