Upsampling artifacts in neural audio synthesis
This addresses a specific technical issue for researchers and practitioners in neural audio synthesis, but it is incremental as it adapts known concepts from computer vision to audio.
The paper tackled the problem of upsampling artifacts in neural audio synthesis, which had been overlooked, and found that nearest neighbor upsamplers can be an alternative to state-of-the-art methods prone to tonal artifacts.
A number of recent advances in neural audio synthesis rely on upsampling layers, which can introduce undesired artifacts. In computer vision, upsampling artifacts have been studied and are known as checkerboard artifacts (due to their characteristic visual pattern). However, their effect has been overlooked so far in audio processing. Here, we address this gap by studying this problem from the audio signal processing perspective. We first show that the main sources of upsampling artifacts are: (i) the tonal and filtering artifacts introduced by problematic upsampling operators, and (ii) the spectral replicas that emerge while upsampling. We then compare different upsampling layers, showing that nearest neighbor upsamplers can be an alternative to the problematic (but state-of-the-art) transposed and subpixel convolutions which are prone to introduce tonal artifacts.