Docling: https://github.com/docling-project/docling

There is a dolphin language model project from Google and Georgia Tech: https://blog.google/technology/ai/dolphingemma/

That's exactly the kind of thing I was hoping people were working on!

As far as I understand, the MMS TTS models are trained from scratch (section 7.1 of [1]), they do not employ any SSL models. So the OmniASR SSL models are not useful here.

What might be interesting is the newly released OmniASR data, because the MMS data, which was used for the MMS TTS, was never released.

Also, the OmniASR can be used to transcribe some untranscribed speech to train a TTS on it.

[1] MMS paper: https://arxiv.org/pdf/2305.13516

You’re completely right, I misremembered. I edited my post.

How is/was the data written to disks? Something like rsync/netcat?

We use the same nginx rust server to do file writes, it's done via web requests

Probably a mix-up with the recently released Huawei model:

https://news.ycombinator.com/item?id=44441447

What is the motivation to move startup from Canada to Germany, if I may ask?

I would be interested to see this reproducibility study.

Could this (or something similar) be found in public access/some libraries?

There is only a single paper that has published a similar derivation but with a critical mistake. To be fair there are many documented examples of how to derive parametric relationships in linkages and can be quite methodical. I think I could get Gemini or 3.5 to do it but not single shot/ultra fast like here.

Telegram?

Doesn't moving air help evaporation anyhow?

If evaporation is possible, yes.