eghad · HackerTrans

5 лет назадdiscuss

Congrats on the launch! Worked on something similar, so rooting for ya. Using Hollywood accent coaches is a great approach, but I'd also look into identifying speech therapists that specialize with helping adults in target regions to 1) get more information for your models and 2) point out which problems can't be corrected in older adults, so you focus on those that can.

eghad

5 лет назадdiscuss

The problem with adapting this to other forms of atypical speech is that their recommendation system likely relies on a catalog the phonemes L2 speakers have issues with (the example most people know is the Japanese "L" "R" swap) so it's much easier to create courses with specific focuses and solutions.

If Google's Project Euphonia [0] is actually still ongoing and they release their dataset/methodology of training models with that sparse dataset I can see your idea as approachable; even accented speech is a tough problem to work on considering how many variants exist worldwide (but their approach looks good!).

[0] https://sites.research.google/euphonia/about/

eghad

5 лет назадdiscuss

I'm even talking vs custom trained models with Kaldi (was working on a startup that was trying to create lessons for public speaking so we could grab enough data to tackle accent remediation/help those with aphasic speech disorders) and again just reiterating, the out of the box performance of Nuance's products are just better than anything else.

Obviously Nuance is more than just speech recognition, but still not sure why people are downplaying how good they were at it.

EDIT: or maybe it's just too prohibitively expensive for people outside of medical/legal fields to know about? And don't get me wrong, I love that things like Talon Voice are widely available for hands free coding, I just hope this means NaturallySpeaking will supplant Windows Dictation.

eghad

5 лет назадdiscuss

And what do you suggest is better? I've worked with nearly every tool (open source and closed) under the sun in medical, industrial, and personal settings and Dragon NaturallySpeaking/Professional was by far the best in terms of accuracy regardless of prosody, accent, background noise, technical terms used, etc.

Personally I think they should've been acquired a decade ago.

eghad

5 лет назадdiscuss

Because the manufacturer of the medicine explicitly said it can't treat the off-label use? And won't sell it to anyone claiming it can.

eghad

5 лет назадdiscuss

I never said zero stakes? There are clear instances where a a 1 or 0 shot transformer can have benefits beyond entertainment--topic modeling and named entity recognition for instance (I'm on the team that believes that human-in-the-loop systems will always outperform solo systems on their own and that GPT-3 alone does not confer any competitive advantage). If you think that chatbots are the only user facing use-case for a transformer, then frankly that's on you falling for the hype surrounding its language generation performance.

OpenAI knows GPT-3 is not sophisticated enough to perform medical diagnosis or analysis (anyone can look at how Watson failed), so it'd never approve such a risky application.

eghad

5 лет назадdiscuss

I mean the OpenAI team would never approve this application for production. It's very clearly stated (in both the article and use case guidelines) that medical diagnosis would be a "high stakes domain" and is unsupported. Frankly, I'm not sure why this result is even notable.

eghad

6 лет назадdiscuss

Abstracts are important (and clearly key in generating these TLDRs), but when it comes to ranking and recommending other papers (not to mention noting whether a new paper has content that can actually push a field forward) an abstract just isn't enough.

eghad

6 лет назадdiscuss

The Koch's political funding goes far beyond just this one PAC fyi. Their funneling of millions into the Federalist Society and its individual causes serving as just one example that has had an outsized negative influence on the health of American democracy.

eghad

6 лет назадdiscuss

Not only that, but the assumptions are poor to begin with. Most skincare products containing sodium hyaluronate are either micronized or in a crosspolymer so that it can penetrate the skin, not HA on its own (something I'd expect a chemist/dermatologist to know). Cursory googling it seems the writer/owner was attaching her name to reddit submissions of scientific articles unrelated to her to game SEO and give her credibility that it it's painfully obvious she doesn't have.

eghad

6 лет назадdiscuss

No, iOS devices can also soft-brick, it's just a general term. Most often people encounter the term in relation to hacking/rooting hardware, like a Nintendo DS/Switch for instance. Hitting up Wikipedia or Trends can help you dig through its usage considering how Google's search algo heavily biases current events wrt relevance.

comments