I've generated a full set of US-English sounds using the voice Matthew, with the fancy "neural" setting, for 16000 and 8000.

https://github.com/avimar/freeswitch-sounds-polly has install instructions.

Once you switch to a polly voice, you can use this script to generate new prompts that instantly match the rest of your prompts.

I've tested with mod_say_en and it sounds decent. I think neural is optimized for actual sentences, and not single words. So perhaps digits and letters are best replaced with the more uniform Matthew-standard (included). I'll try to test it more another time. Feedback welcome.

If people want 24000 I can generate as ogg and then convert with sox, I guess? Polly doesn't support above 24000hz. 

The script can read a .xml file but it's currently set a) to only do one section at a time and b) to request pcm at 16000 and use sox to downsample to 8000.

If someone wants to create other sets, you can grab the phrases file from  https://github.com/signalwire/freeswitch/tree/master/docs/phrase and pick a voice from https://docs.aws.amazon.com/polly/latest/dg/voicelist.html (you'll have to switch the XML/JSON importer, it defaults to english)

-There's neural voices available for English (British) (en-GB), Portuguese (Brazilian) (pt-BR) (phrases), and US Spanish (es-US).

-There's regular voices available in lots of others languages, we have phrases for these: German (de-DE), French (fr-FR), Portuguese (European) (pt-PT), Russian (ru-RU), Chinese, Mandarin (cmn-CN).

Email me or file an issue if you need help.