anyone have good ideas for how to get an agent to spell out a name properly? all the TTS models i find are bad at spelling words, especially in other languages like Spanish

other languages, not sure, but I think rime and cartesia have some good ways to approach this. i've found that to get things to be natural, you sometimes want to introduce your own logic of like pausing after groups of 3 letters. e.g. "S O L A R" is easier to hear than "S O L A R"

cool idea, i'll give that a try. the TTS models still are very weird about spaces and not consistent

https://rime.ai/resources/spell-function this is worth a try! i actually suggested this format to them because I was instructing my agent at the time to just output "It's spelled spell(solar)" and then I was doing some regex stuff afterwards to turn it into "S <break/> O <break/> ..."

I implemented something like that before + used https://github.com/keredson/wordninja + a bunch of rules