JenGAN: Stacked Shifted Filters in GAN-Based Speech Synthesis

Non-autoregressive GAN-based neural vocoders are widely used due to theirfast inference speed and high perceptual quality. However, they often sufferfrom audible artifacts such as tonal artifacts in their generated results.Therefore, we propose JenGAN, a new training strategy that involves stackingshifted low-pass filters to ensure the shift-equivariant property. This methodhelps prevent aliasing and reduce artifacts while preserving the modelstructure used during inference. In our experimental evaluation, JenGANconsistently enhances the performance of vocoder models, yielding significantlysuperior scores across the majority of evaluation metrics.