Currently running some finetuning experiments on non-verbal sounds to teach TTS how to laugh. I have had some success to add the necessary tags and tokens to multiple systems, but assembling the necessary dataset with sufficient quality is hard.