My interests are in local inference, so a service like Amazon Polly doesn't fit my needs.

I haven't tried anything so far that I actually liked so my answer to "what's the best quality I have found?" is none.

The reverse though, OpenAI's Whisper for speech to text has been amazing, far exceeding my expectations.

Having just looked again I found a list at huggingface that I will be poking through in my spare time soon.

https://huggingface.co/models?pipeline_tag=text-to-speech&so...