Announcing the Initial Release of Mozilla’s Open Source Speech Recognition Model and Voice Dataset

Announcing the Initial Release of Mozilla’s Open Source Speech Recognition Model and Voice Dataset

At Mozilla we’re excited about the potential of speech recognition. We believe this technology can and will enable a wave of innovative products and services, and that it should be available to everyone.
And yet, while this technology is still maturing, we’re seeing significant barriers to innovation that can put people first. These challenges inspired us to launch Project DeepSpeech and Project Common Voice. Today, we have reached two important milestones in these projects for the speech recognition work of our Machine Learning Group at Mozilla.
I’m excited to announce the initial release of Mozilla’s open source speech recognition model that has an accuracy approaching what humans can perceive when listening to the same recordings. We are also releasing the world’s second largest publicly available voice dataset, which was contributed to by nearly 20,000 people globally.

Read the full article:
blog.mozilla.org/blog/2017/11/29/announcing-the-initial-release-of-mozillas-open-source-speech-recognition-model-and-voice-dataset/

discuss

Other urls found in this thread:

voice.mozilla.org/
itunes.apple.com/us/app/project-common-voice-by-mozilla/id1240588326?mt=8
twitter.com/AnonBabble

At least it wasn't written in rust.

Probably will be used for spying somehow.

voice.mozilla.org/
Downboat all Niggers/Poo in the Loos/Females tbh

In a way v57 was the best thing that ever happened to Firefox. Now that I've freed myself, don't have to give two shits about what Mozilla is up to anymore.

epic. sage negated btw
posted with firefox quantum v58.0b7 (64-bit)

itunes.apple.com/us/app/project-common-voice-by-mozilla/id1240588326?mt=8
some apple users are not a lost cause

voice.mozilla.org/
you can also do it with out the app. i have the inclination to read off some quality material into it but I really don't want to give away my voice, maybe with a fuck with the voice.

"Would you like a stamp?"

It's disappointingly rare for machine learning projects (especially authors of a paper) to properly provide all of the implementation source code, the full training set, applicable processing and tokenization scripts, and pretrained models. Mozilla seems to have actually set their priorities straight for once in creating a fully FOSS end-to-end speech recognition system.

What, and shit up the one project Mozilla didn't completely fuck up? I was looking forward to using this training set myself.

are you happy being a goy?

After all the shit Mozilla has been up to, i'm skeptical, but who knows? It could be a good project.

you forgot to call me a soyboy cuck

I bet 10b dollarydoos they're all secretly sent to gawgle

I bet it's permissively licensed so google can get even more data without paying and keep all theirs proprietary.

CC-BY-SA or bust

...

this is probably the rustfag

Who is funding Mozilla to work on these silly pet projects?

(((Google))) shill/thrall

(((Google))) shill/thrall

(((Google))) shill/thrall

(((Google))) shill/thrall

Issue is that your voice goes into a public database... Anyone cal troll through it and get your voice.

I wonder how exploitable this is. Can you steer it in a bad direction with bad samples (ie. giberish, pajeet speak).
Can you use this as a means of leaving behind coded messages?

mozilla can tie the voice to an apple account, which is tied to a real name. their anonymity is always a lie.
i don't know how useful even having some random lemming's voice with a full name is though.

that would be interesting, i wonder if you could encode shit into the audio stream, and if they would be able to process it out. if you wanted to sabotage the database you encode all manner of illegal or questionable shit. i don't really see why that's useful either though. fuck mozilla and all that but huge sample's of voices with transcriptions for free is useful for everyone.

If the system is even half competent then it's already preloaded with prediction software. If it predicts that X word will contain Y sounds, and it gets Z instead then it will probably reject it without telling you, and you'd just be wasting hours giggling to yourself but not actually accomplishing anything.

I found one nonsense sample and some clearly wrong ones. Why waste CPU cycles on verifying stuff that will get verified by volunteers for free.

I am the system administrator.
My voice is my password.
Verify me.

this is why you never use any form of biometric data for passwords, ever. usernames maybe if your too fucking lazy to type it in

Stopped reading there.

Biometrics are for identification, not authorization.

...

why WASN'T it written in rust? wtf!

correct. i made this thread btw

They wanted to appeal to your tastes, user.

Steganocrypt some CP inside... Then half a year later leak the decryption method to media outlets. What a nice idea, user!

Someone once encoded CP links into the bitcoin blockchain in some transactions' coinbase parameter but nobody's calling for a ban on that basis.

Well, it's links, not CP itself. Those links are probably dead by now. Many file hostings maintain hashes of known CP files. Arguably, those hashes may be considered "links", because that's precisely how BitTorrent + DHT work. Posting a link to CP is not illegal, but storing it even in an encrypted form may be, depending on your jurisdiction.

...

I think it is in some countries.
I might be thinking of linking to copyrighted material

...

I want to start a file hosting service where would I get this hash DB? this is the shit no one wants to talk about or admit to.

I can't wait for an actually useful speech to text engine, I have tens of thousands of hours of podcasts and speeches in need of transcription.

Most CP is copyrighted. I doubt you'd find many people willing to try enforcing their copyright on it, though.