Microsoft's latest models are very, very good at this.

Try the demos here (and be sure to experiment with the "speaking style" parameter box, and try out the "(Neural) Preview" voices for US English): https://azure.microsoft.com/en-us/services/cognitive-service...