You have to anyways because different TTS models say things different. So we only offer a few voice choices because the dataset we’ve made is specific for those voices and the way they handle different symbols