Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Meta AI announces Massive Multilingual Speech code, models for 1000+ languages

github.com

Post reply on HN
Print view View original

111–120 of 232 posts

Re: Meta AI announces Massive Multilingual Speech code, models for 1000+ languages

#111

Post by sneak »

kunwon1 wrote: post #92
sneak wrote: post #77

Earlier quoted context omitted.

This is what containers solve. Don't waste time manually installing things. Store state in a database via the app on a different host.

Speaking as someone who has encountered similar difficulties, this response has strong 'Draw the rest of the owl' vibes

Speaking as someone who has solved these difficulties hundreds of times, "draw the rest of the owl" doesn't tell you the specific things to google to get detailed examples and tutorials on how millions of others have sidestepped these repeated issues.

Re: Meta AI announces Massive Multilingual Speech code, models for 1000+ languages

#112

Post by RheingoldRiver »

mkl wrote: post #89
neom wrote: post #7

Earlier quoted context omitted.

"According to the World Atlas of Languages' methodology, there are around 8324 languages, spoken or signed, documented by governments, public institutions and academic communities. Out of 8324, around 7000 languages are still in use." https://en.wal.unesco.org/discover/languages

Most are at risk of extinction. "half of the languages spoken today have fewer than 10,000 speakers and that a quarter have fewer than 1,000 speakers" ( https://en.wikipedia.org/wiki/Language_death ). "Today, on average, we lose one language in the world every six weeks. There are approximately 6800 languages. But four percent of the population speaks 96 percent of the languages, and 96 percent of the population spea…

Wow, preserving almost-dead languages sounds like something that LLMs would be pretty appropriate for, right? We would primarily need as large a body of written text translated into both a "known" language and the dying language as possible.

Re: Meta AI announces Massive Multilingual Speech code, models for 1000+ languages

#113

Post by jimmySixDOF »

simonw wrote: post #41
crakenzak wrote: post #2

Code: https://github.com/facebookresearch/fairseq/tree/main/exampl... Blog Post: https://ai.facebook.com/blog/multilingual-model-speech-recog... Paper: https://research.facebook.com/publications/scaling-speech-te... Languages coverage: https://dl.fbaipublicfiles.com/mms/misc/language_coverage_mm...

I loaded the language coverage into Datasette Lite and added some facets here: https://lite.datasette.io/?json=https://gist.github.com/simo... Here's how I did that: https://gist.github.com/simonw/63aa33ec827b093f9c6a2797df950... Here are the top 20 represented language families: Niger-Congo 1,019 Austronesian 609 Sino-Tibetan 288 Indo-European 278 Afro-Asiatic 222 Trans-New Guinea 219 Otomanguean 149 Nilo-Saharan 13…

All this great work and the prior sota translations and Meta still only accepts Nort American english voice control in their VR equipment lol.

Re: Meta AI announces Massive Multilingual Speech code, models for 1000+ languages

#114

Post by sdenton4 »

tmpz22 wrote: post #83
crazygringo wrote: post #56

Earlier quoted context omitted.

They're intended for researchers/professionals not consumers, and I'm not sure how a video is going to be helpful? And the issue with a live demo is that these are resource-intensive, they're not just webpages. It's an entire project to figure out how to host them, scale them to handle peaks, pay for them, implement rate-limiting, and so forth. For the intended audience, download-and-run-it doesn't seem like an issue…

Its so weird to me that they'd do 99% of the effort and just skip past the 1% of work to provide a dumbed-down summary and instructions for broader appeal. Clearly these are released in part for public-relations and industry clout. Don't get me wrong they published this, it took a ton of work, they didn't have to do it. But its ultimately a form of gatekeeping that seems to come straight out of Academia. And honestly…

A good research team is a handful of people working on a focused set of questions.

A good product team is probably around a dozen people minimum? Especially if you need to hit the quality bar expected of a release from a BigCorp. You've got frontend, UX, and server components to design, in addition to the research part of the project. The last real app I worked on also included an app backend (ie, local db and web API access, separate from the display+UX logic) and product team. Oh yeah, also testing+qa, logging, and data analysis.

And after all that investment, God help you if you ever decide the headcount costs more than keeping the lights on, and you discontinue the project...

Public app releases are incredibly expensive, in other words, and throwing a model on GitHub is cheap.

Re: Meta AI announces Massive Multilingual Speech code, models for 1000+ languages

#115

Post by gcr »

hmoodie wrote: post #65
simonw wrote: post #63

Earlier quoted context omitted.

That's a related complaint: everyone continues to insist on releasing papers as PDFs, ignoring the fact that those are still pretty nasty to read on a mobile device. Sure, release a PDF (some people like those), but having an additional responsive web page version of a paper makes research much more readable to the majority of content consumption devices. It's 2023. I'll generally use https://www.arxiv-vanity.com/ to…

Would you like a back rub while I'm at it?

fwiw, i know the topic of the thread has deviated, but i share the frustration about reading PDFs on my device. In my case, it's an accessibility issue - I can't see well, so zooming with reflow would make my life materially better since I'd be able to read research papers on my morning commute.

Sometimes users have needs that may seem superfluous and beg for a snarky reply, but there are often important reasons behind them, even though they may not be actionable.

I'd pay $x000 for an app that does some sort of intelligent pdf-to-epub conversion that doesn't require human-in-the-loop management/checking.

Re: Meta AI announces Massive Multilingual Speech code, models for 1000+ languages

#116

Post by barking_biscuit »

crazygringo wrote: post #56
m3kw9 wrote: post #36

The problem with all these model releases is they have no demos or even video of it working. It’s all just download it and run it, like it’s an app.

They're intended for researchers/professionals not consumers, and I'm not sure how a video is going to be helpful? And the issue with a live demo is that these are resource-intensive, they're not just webpages. It's an entire project to figure out how to host them, scale them to handle peaks, pay for them, implement rate-limiting, and so forth. For the intended audience, download-and-run-it doesn't seem like an issue…

We're all research professionals now.

Re: Meta AI announces Massive Multilingual Speech code, models for 1000+ languages

#117

Post by KaoruAoiShiho »

jokethrowaway wrote: post #72
KaoruAoiShiho wrote: post #52

Earlier quoted context omitted.

Centuries really? I think Meta's going to be just on time.

I think we may skip the whole VR experience and just have robots instead of virtual friends

That's a large overlap rather than sequential.

Re: Meta AI announces Massive Multilingual Speech code, models for 1000+ languages

#118

Post by barking_biscuit »

idiotsecant wrote: post #42
Hamuko wrote: post #15

Earlier quoted context omitted.

It's not VR, it's Metaverse; VR might involve experiences that are not bullshit.

The metaverse, if we define it as an information layer that exists in parallel with the physical 'stuff' universe and is a seamless, effortless, and essential part of what we experience as reality will be an enormous part of our future. Meta might just be a few centuries ahead of the curve, which is just as bad as being a few centuries late.

If you define it like that, it's just the internet.

Re: Meta AI announces Massive Multilingual Speech code, models for 1000+ languages

#119

Post by sarabande »

I'm trying to use this on a 3M mp3 file to test ASR with language code deu, CPU only, and I keep getting this error -- are there limits to the MMS inference?

  File "fairseq/data/data_utils_fast.pyx", line 30, in fairseq.data.data_utils_fast.batch_by_size_vec
    assert max_tokens 
    process(args)
  File "/home/xxx/fairseq/examples/mms/asr/infer/mms_infer.py", line 44, in process
Post reply on HN