I've just started to use AWS Polly in some demos, but the speech synthesis quality feels 2-3 years old at least :/