[TTS] riva model support, what to choose

[itzsimpl]

We're experimenting with Fastpitch+Hifigan TTS on Slovene language, and the results are quite good. Pronunciation is very good, but we have issues with certain artefacts present in the voiced output, so we're investigating other models as well. From the collection of models available in NeMo we're currently looking into VITS and RadTTS.

VITS because an implementation from another repo trained on the same dataset does not show these artefacts. RadTTS because Riva has gained support for it in 2.10.0 (support for Fastpitch+Hifigan, has been there for a long time).

Now the questions are; are there any plans that Riva will support VITS as well? Are there any plans that NaturalSpeech will get implemented in NeMo (#6746) and get support in Riva later on? What is the current status of the training scripts for VITS and RadTTS? Are there any other TTS models worthy of looking into?