[ipsum2]

Whatever they're using for text to speech is rough. Probably using an open source model. The one used in OP (Google's) is a lot more listenable.