However, traditional vocoders and modern neural vocoders (like WaveNet or HiFi-GAN) suffer from two distinct problems: