[nshm]

It is explainable given that there was a single developer working on TTS. It is hard to compete with big academic teams/industry players this way.

I also believe Mozilla team was restricted by a lack of computing resources. They had just a single 8GPU server or so.

[posguy]

Said 8 GPU server was consistently in use for Mozilla DeepSpeech (now renamed Mozilla STT) in training models. Its impressive how far Mozilla got considering how limited their resources were.

[qchris]

This is an area that I find unbelievably frustrating. A lack of computing resources in the current day is kind of insane. You can buy an 8GB GPU for <$1000. Even with the rest of the costs, the cost of hardware like this is a drop in the bucket when your main office is housed in Mountain View! Especially on a project that ends up being public-facing, these are missed opportunities where a little can go a long way.

[nmstoker]

I take your point but according to the release details on the repo it was not 8Gb on one card but a server with 8 cards, each a Quadro RTX 6000 with 24Gb, and they're around £4k each currently, so the cost of the GPUs alone is £32k

https://github.com/mozilla/STT/releases/tag/v0.8.2

[qchris]

Ah, I see-- not an 8GB, 1-GPU server, but an 8-GPU server. That does make a bit of a difference, changing the cost from a new workstation to functionally a piece of capital equipment. Still, I'm not sure that my point about equipment costs falls short--even at (call it) $40K, you're probably talking less than 3 months of the company's all-in cost for the developer themself, amortized over multiple years.

[panpanna]

We need a SETI@home approach to open source AI models.

Only then we can break our dependency on Google and Facebook - and Mozilla for that matter.

[nshm]

It is going on, see https://github.com/learning-at-home/hivemind and https://github.com/bytedance/fedlearner