Regarding the distortion at the beginning and end of phrases, this is a known characteristic of RVC models. However, there are several improvements we can apply to reduce it.