Last time this came up the options were the Mozilla TTS service or do it locally with tensorflow lite and DeepSpeech.