The first example sounds like a cry for help.

Some of them have tone wobbles which iirc was more common in early TTS models. Looks like the huge context window is really helping out here.