|
|
Subscribe / Log in / New account

Mozilla releases its speech-recognition system

Mozilla has announced the initial releases from its "Project DeepSpeech" and "Project Common Voice" efforts. "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."

to post comments

Mozilla releases its speech-recognition system

Posted Dec 4, 2017 18:22 UTC (Mon) by dc123 (guest, #117760) [Link] (29 responses)

Glad to hear that Firefox is finally fixed and they can focus on other stuff!

Mozilla releases its speech-recognition system

Posted Dec 4, 2017 18:38 UTC (Mon) by atai (subscriber, #10977) [Link]

the people working on this may not be the same people working on Firefox

Mozilla releases its speech-recognition system

Posted Dec 4, 2017 18:47 UTC (Mon) by Cyberax (✭ supporter ✭, #52523) [Link] (25 responses)

Well, FireFox released a massive speed improvement with FF57. It's now competitive with Chrome in speed, and it's going to become even better.

Speech synthesis and recognition are two areas that are really lacking any good OpenSource solutions. And it's not easy or feasible to simply write one by yourself, you need a lot of model data and training.

Mozilla releases its speech-recognition system

Posted Dec 4, 2017 20:26 UTC (Mon) by flussence (guest, #85566) [Link] (23 responses)

It's getting there. I'll stick with Chromium until they make it a bit less of a trash fire (::-webkit-scrollbar flickers constantly, entire pages redraw randomly while trying to navigate them, can't use the browser sync any more because they've started using its email addresses as a spam list) but it's nice to see some part of Mozilla is actually trying.

Mozilla releases its speech-recognition system

Posted Dec 4, 2017 22:01 UTC (Mon) by Frogging101 (guest, #113180) [Link] (15 responses)

> they've started using its email addresses as a spam list

Huh? What do you mean by this?

Mozilla releases its speech-recognition system

Posted Dec 4, 2017 22:35 UTC (Mon) by flussence (guest, #85566) [Link] (14 responses)

On October 14 my fxsync-catchall email address got sent a mass-mailed advertisement for Pocket. I had not given Mozilla permission to use my address in this way when I signed up, and they provide no opt-out in the account settings for such marketing junk, so I forwarded the message to spamcop and deleted the account entirely.

Mozilla releases its speech-recognition system

Posted Dec 4, 2017 23:00 UTC (Mon) by mjblenner (subscriber, #53463) [Link]

> and they provide no opt-out in the account settings for such marketing junk

Firefox account -> Manage account -> Communication preferences ?

I don't remember a tickbox when I signed up, but I'm unsubscribed to "Latest news", and haven't received any non-account emails from Mozilla.

Mozilla releases its speech-recognition system

Posted Dec 4, 2017 23:13 UTC (Mon) by Frogging101 (guest, #113180) [Link] (12 responses)

On accounts.firefox.com I see a "Communication preferences" section, under which it says:

> We may send you emails related to your Firefox Account.

> Additionally, you can choose whether to receive the latest news about Mozilla and Firefox.

And then a blue subscribe button. It doesn't look like I'm subscribed, and I know I never explicitly unsubscribed, so I don't think I ever was. I believe I created my account quite some time ago (4+ years?), however.

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 6:57 UTC (Tue) by flussence (guest, #85566) [Link] (11 responses)

There aren't that many visible options to search through on Firefox Accounts and I'm absolutely certain there was nothing in there I'd accidentally left checked. If it were that kind of user error I'd feel deservedly stupid for a while and get over it.

But no, they abused the sanctity of my inbox (for some inane rubbish about a junk toolbar that can't be removed but they know everyone hates), and that's a cardinal sin up there with the likes of Sony trojaning audio CDs. I won't be forgetting it for a long time.

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 7:19 UTC (Tue) by andrewsh (subscriber, #71043) [Link] (9 responses)

You seriously need to calm down.

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 9:17 UTC (Tue) by donbarry (guest, #10485) [Link] (8 responses)

Don't you think we should hold organizations within the free/libre model to higher standards? I do, because the standards were born of the community, and entrusted to a degree to groups like this.

I've been a strong critic of Mozilla's corporate management, though this speech-recognition contribution is a substantial gift and will enable significant improvements in free software accessibility while using free software.

But I agree, I don't give a pass to Mozilla for its cavorting with the likes of Pocket, and if I used their sync services (I don't, since I don't trust them) I'd be infuriated to learn they were selling me out as the same piece of meat that 'everyone else' does.

It's been a long slippery slope, and we're far from the bottom.

Alexander Pope:
Vice is a monster of so frightful mien
As to be hated needs but to be seen;
Yet seen too oft, familiar with her face,
We first endure, then pity, then embrace.

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 12:16 UTC (Tue) by TheLessThanAmazing (guest, #119480) [Link] (3 responses)

> I don't give a pass to Mozilla for its cavorting with the likes of Pocket

Cavorting? Pocket has been wholly owned by Mozilla since February.

Mozilla releases its speech-recognition system

Posted Dec 6, 2017 19:44 UTC (Wed) by lsl (subscriber, #86508) [Link] (2 responses)

Has the source code been opened up yet, like announced back in February? Can't find it, neither the server-side code nor the mobile apps.

Mozilla releases its speech-recognition system

Posted Dec 6, 2017 20:18 UTC (Wed) by Cyberax (✭ supporter ✭, #52523) [Link]

Mozilla releases its speech-recognition system

Posted Dec 7, 2017 9:23 UTC (Thu) by rillian (subscriber, #11344) [Link]

Pocket announced this week that they were publishing source for three components related to non-Firefox browser extensions. Firefox integration was started as a public project. AFAIK source for the Android and iOS clients has not been published, nor has the server.

The announcement was at the weekly public status meeting, around 20 minutes in.

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 12:19 UTC (Tue) by tajyrink (subscriber, #2750) [Link] (3 responses)

You don't need to "trust" Mozilla for their sync services, that's the point of using sync services where the source code is open and you (the user) hold the encryption keys, unlike in other products.

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 12:58 UTC (Tue) by pizza (subscriber, #46) [Link] (2 responses)

I might add that I've run my own "mozilla sync" service for several years now. No data at all held on Mozilla's servers.

Mozilla releases its speech-recognition system

Posted Dec 15, 2017 7:53 UTC (Fri) by flussence (guest, #85566) [Link] (1 responses)

How's that working out? I knew someone who set one up back when it was still an open system, but they gave up trying to keep it running after a while.

It's something I've often wondered about doing myself, but Mozilla have thoroughly whitewashed the browser UI and their website of any hint that users ever had a vote in the matter. Maybe my pockets aren't deep enough to know.

Mozilla releases its speech-recognition system

Posted Dec 15, 2017 12:49 UTC (Fri) by pizza (subscriber, #46) [Link]

The original v1.x ffsync service was entirely self-contained, but it turned out that users were too special to cut-n-paste the authentication token to other devices, so v2. went to more traditional login system.

I stuck with v1 until the browser dropped support for it altogether (ff v55 I think), because of the royal PITA it was to install the authentication backend that the v2 sync service relied upon.

But now I have a v2 ffsync instance running, but it's talking to the official firefox authentication servers. [1] Took something like ten minutes to set up, including changing the preference on the browser clients to point at my sync instance..

[1] The only information that mozilla stores about me is my email address and a unique password. And the timestamps when I authenticated each individual browser. The actual sync service is entirely self-contained.

Mozilla releases its speech-recognition system

Posted Dec 19, 2017 14:53 UTC (Tue) by knobunc (subscriber, #4678) [Link]

> toolbar that can't be removed but they know everyone hates

Right click > Remove from Address Bar.

-ben

Would be nice to have bug citations instead of bellyaching

Posted Dec 4, 2017 22:24 UTC (Mon) by sdalley (subscriber, #18550) [Link] (6 responses)

> webkit-scrollbar flickers constantly
Not on mine, it doesn't.

Would be nice to have bug citations instead of bellyaching

Posted Dec 4, 2017 22:37 UTC (Mon) by flussence (guest, #85566) [Link] (5 responses)

To be fair, this may be because Intel's drivers are infamously bad.

Would be nice to have bug citations instead of bellyaching

Posted Dec 5, 2017 1:36 UTC (Tue) by roc (subscriber, #30627) [Link] (2 responses)

One reason vendors get away with bad drivers is that users tend to blame the application first. After all if Firefox renders incorrectly but Chrome is OK, it must be Firefox's fault right?

Would be nice to have bug citations instead of bellyaching

Posted Dec 5, 2017 8:08 UTC (Tue) by flussence (guest, #85566) [Link]

Oh, but Chromium is *even worse* on that system - it has a habit of causing complete display freezes when loading heavy websites (e.g. scrolling through a twitter feed with embedded videos), sometimes several minutes long. To add insult on top of that, it usually pops up a hung tab quit button after it recovers. That was hardly the only thing, but it's the main reason I bothered to try Firefox 57 at all.

...and the worst complaint I've come up with is that it doesn't correctly render a badly written web app that uses proprietary features from someone else's browser. Maybe I should stop whining and give it a chance now.

Would be nice to have bug citations instead of bellyaching

Posted Dec 5, 2017 15:02 UTC (Tue) by daenzer (subscriber, #7050) [Link]

We get our fair share of bug reports about application issues reported against driver components as well.

Would be nice to have bug citations instead of bellyaching

Posted Dec 5, 2017 12:19 UTC (Tue) by TheLessThanAmazing (guest, #119480) [Link] (1 responses)

I'm running Firefox 57.0.1 x86_64 on Fedora 27 with Intel graphics (i5 3570 with HD Graphics 2500 to be exact) and have never seen any such flicker.

Would be nice to have bug citations instead of bellyaching

Posted Dec 5, 2017 20:05 UTC (Tue) by zlynx (guest, #2285) [Link]

At least one case of Intel flickering bugs is PSR (Panel Self Refresh) which saves a lot of power but was quite buggy for a long time. My laptop is a Haswell CPU and still has PSR bugs. If I force enable it, under gnome-shell using Wayland, I get a nice savings from 12W to 8W idle use but there are many things that flicker annoyingly.

The PSR enable and it being used from user-space all depend on a complicated set of rules about compatible versions of hardware and software, so it is very possible to have one Intel GPU flicker and another one not flicker. Or even the same hardware but depending on Ubuntu vs Fedora. Or X.org vs Wayland.

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 15:35 UTC (Tue) by ajmacleod (guest, #1729) [Link]

In my testing, FF57 runs at exactly the same speed as Pale Moon whilst consuming vastly more RAM - I certainly won't be switching back any time soon (ever, I expect.) Then again, I've not found Chrome to be faster than FF either - not that browser speed has ever been much of an issue for me.

Speech recognition is definitely not something I want, anywhere, ever (and works abominably every time I've ever used it) but I can see the benefits to others from an accessibility point of view; it's a slightly less pointless diversion than Mozilla have been prone to wasting resources on over the past few years.

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 1:16 UTC (Tue) by roc (subscriber, #30627) [Link] (1 responses)

Actually Firefox needs a quality speech recognizer to implement the standard WebSpeech API that provides speech recognition to Web applications. So this work might well become part of Firefox.

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 1:24 UTC (Tue) by roc (subscriber, #30627) [Link]

Of course Mozilla did have the option of delegating speech recognition to the underlying OS, so that things work reasonably well for Mac/Windows/Android users and Linux users are left high and dry. Would you have been less snarky about that?

Mozilla releases its speech-recognition system

Posted Dec 4, 2017 18:59 UTC (Mon) by karkhaz (subscriber, #99844) [Link] (9 responses)

Wow, how exciting! I wonder if the Librem folks could use this to create a compelling Siri alternative for their upcoming smartphone (and desktop too, why not).

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 5:59 UTC (Tue) by eru (subscriber, #2753) [Link] (1 responses)

I don't care about Siri-like assistants, but perhaps this could enable the creation of a good dictation-taking program, perhaps as a LibreOffice extension: Open LibreOffice and start speaking to it to get a reasonable first draft. THAT would be useful.

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 11:14 UTC (Tue) by nix (subscriber, #2304) [Link]

That sort of thing is useful for people with really bad RSI as well. At the moment there is no free option for decent speech recognition. (Obviously using speech recognition to code is a total drag because of all the punctuation, but rather a total drag and a lot of macro hackery than total agony.)

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 13:44 UTC (Tue) by ledow (guest, #11753) [Link] (6 responses)

Given that I spent an instructive half-hour in a car with some friends yesterday, where we were actually all loading up various smartphones, sat nav apps, car voice interfaces, etc. and NOBODY successfully managed to get to the point where we could issue an instruction along the lines of:

"Navigate to Edgware Station"

and get ANYTHING useful back whatsoever, I'd rather we really didn't waste time on voice recognition at all, certainly trying to make dictation engines.

We were literally unable to make the simplest commands, spoken in the most normal manner, in the quietest of cars (not even moving for the first ten minutes but then we gave up and headed the way we knew we would have to start), using everything we had available and the range of voices we had get close. And we were honestly trying, because we didn't know the way and wanted a quick start.

I was also once in a seminar for teachers that suggested they could dictate their school reports via the Dragon iPad apps. Obviously the "demo" was amazing. Real-life trial? Really not worth the effort of even loading up a trial app. Abandoned the idea with a matter of minutes.

Speech recognition really has a LONG way to go, but people somehow ignore the simplest of failures from it, and I'm not sure tying it into every product is at all a sensible idea, accessibility-wise, either. For a start, almost NOTHING understands my voice, no matter how careful I am. It would literally lock me out of services if we all start going that way.

Though I agree there are uses, heavy use of voice recognition isn't a part of anyone's life that I know. And in fact on those automated cinema lines, from a very limited range of possibilities, most people I know who've tried have just given up and gone online.

A speech recognition project will always be necessary. But tying it into the browser? That's just an creep into things that browsers shouldn't have to contend with at all.

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 23:01 UTC (Tue) by madscientist (subscriber, #16861) [Link]

Interesting. I was driving to the airport on Sunday and realized I hadn't checked traffic. I said to my phone (Nexus 5, so not even getting updates anymore), "directions to XYZ airport" and google maps popped it up and off we went. I didn't even have to try twice. I almost never use voice recognition mostly because I'm just not used to it and don't think about using it. But it works moderately well with straightforward requests, whenever I do try it.

Mozilla releases its speech-recognition system

Posted Dec 6, 2017 5:33 UTC (Wed) by Blortuga (guest, #120092) [Link] (4 responses)

I did QA for a commercial speech recognition SDK; years ago.

Microphone quality is key. (in addition to the software's recognizer algorithm, the parameters, the model, etc).

And smartphone mics, are of varying quality, with varying amounts of pocket-lint crammed into them. And also: AFAIK; smartphones don't have onboard CPU power capable of doing speech recognition, and they're still mostly sending an encoded sample to a server, where the recognition is done, and the text result is sent back - poor network connections can also cause problems.

But anyway - I do not know why smartphone manufacturers do not step up their game with microphone quality. When you're charging $1000 for hardware, a good quality microphone shouldn't be that hard, considering it's part of the primary function of the device, as a phone.

That's my take on why smartphone voice recognition performance is often so "crappy", 8 or 9 years after it first came out (Siri).

Mozilla releases its speech-recognition system

Posted Dec 6, 2017 16:13 UTC (Wed) by khim (subscriber, #9252) [Link]

But anyway - I do not know why smartphone manufacturers do not step up their game with microphone quality.
Because that's not what users demand?
When you're charging $1000 for hardware, a good quality microphone shouldn't be that hard, considering it's part of the primary function of the device, as a phone.
Device whose primary function is voice today are sold for $30 if not for $15. Not much you could do to microphone quality if your whole device in that price range. $1000 devices have NEVER had voice as their "primary function". 10 years ago that were SMS, today it's WhatsApp and camera, but not voice. And even if they DO work on voice all enhancements are not in quality of sound but in the ability to use it when someone nearby is speaking loudly, etc. Thus 2-3 "ambient" mics and clever algorithms, not high-quality mics...

Mozilla releases its speech-recognition system

Posted Dec 7, 2017 1:12 UTC (Thu) by thomasg (guest, #114185) [Link] (2 responses)

Apple shipped Siri (after they acquired it in 2010) starting as beta in the end of 2011, so it actually only has been 6 years.
It's also a stretch to call it the first smartphone based voice recognition, there have been others before that; though they usually weren't remotely as good - but also didn't have the tens of millions of dollars DARPA funding that enabled the SRI (that's where the name Siri comes from) to develop it.
Apple was just quick enough to purchase SRI before they released their android app.

Voice recognition on phones was in fact a thing a decade earlier, when Nokia rolled it out a decade earlier on the 3310 in the year 2000. This was a time when 3G devices did not yet exist, mobile data was limited to at most 64 Kilobits per second, and the voice recognition was done on board with a below 200 MHz ARM7 CPU made by TI (which likely also produced the voice recognition feature, based on their DSPs, though I can't find any solid data about that).

Sorry to go out on a correction-rant on you, it just annoys the hell out of me when history is rewritten, again and again, to paint Apple as the creative minds behind features, when in fact they rarely invented anything and were just very smart in acquiring promising technology and delivering solid products implementing it.

Mozilla releases its speech-recognition system

Posted Dec 7, 2017 6:45 UTC (Thu) by gioele (subscriber, #61675) [Link] (1 responses)

> Voice recognition on phones was in fact a thing a decade earlier, when Nokia rolled it out a decade earlier on the 3310 in the year 2000.

The 1999 model of the Philips Genie already had voice recognition in multiple EU languages for basic tasks (worked decently) and for contacts (worked decently with first names, it has never matched a surname in my experience).

PS: The first 3 revisions of the Philips Genie had the best phone interface ever made. Nokia took years to catch up with that.

Mozilla releases its speech-recognition system

Posted Dec 7, 2017 17:06 UTC (Thu) by thomasg (guest, #114185) [Link]

Interesting. Never came across one of those - and can't even find something remotely resembling hardware specs.
Thanks!

Mozilla releases its speech-recognition system

Posted Dec 4, 2017 20:29 UTC (Mon) by SEJeff (guest, #51588) [Link]

Home Assistant (https://home-assistant.io/) integrates with snips.ai specifically due to it being on-premise only. I wonder how long it will take them to integrate with a service for this.

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 5:18 UTC (Tue) by pabs (subscriber, #43278) [Link] (3 responses)

I wonder why they didn't make use of VoxForge recordings:

http://www.voxforge.org/

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 7:48 UTC (Tue) by Cyberax (✭ supporter ✭, #52523) [Link] (2 responses)

Because Mozilla now has at least an order of magnitude more data than Voxforge?

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 10:15 UTC (Tue) by eru (subscriber, #2753) [Link] (1 responses)

Because Mozilla now has at least an order of magnitude more data than Voxforge?

Maybe in English. Voxforge seems to support several other languages as well. (But not mine, it seems.)

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 10:19 UTC (Tue) by Cyberax (✭ supporter ✭, #52523) [Link]

Check the sample counts. Other languages have even less data.

I was excited that it had Russian support, but it turns out that there is barely any data there. Not useful for AI training, for Russian it's still better to use public domain movies from USSR with subtitles.

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 15:36 UTC (Tue) by smurf (subscriber, #17840) [Link] (1 responses)

… for which languages?

Mozilla releases its speech-recognition system

Posted Dec 5, 2017 22:23 UTC (Tue) by edgewood (subscriber, #1123) [Link]

If you have to ask, it's English only.


Copyright © 2017, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds