Status of Speech Recognition API

[Robert Jones]

Could someone from the Google Speech Recognition API team give us an update on the current status of that API?

Specific questions that I have are:

Is there any difference in functionality between regular Chrome and Chrome Canary?

How much of the W3C api spec is currently implemented ? (I realize that the spec is not totally finalized)

Is there any way to get info on the state of a current connection between the client and Google speech recognition server, other than interim results?

Most of the time this works fine, occasionally I get a 'network error' and then some times it doesn't appear to do anything (even with seemingly clear speech) - I don't have a way to know where that problem lies.

I'm thinking of a log message something like '57 bytes sent to server'...

Are there any 'accuracy' figures for recognition in each language ? I realize that is difficult to define - how well does speech recognition work in terms of % correct words for a given speaker and % correct for a given text spoken by a sample population, etc... are some languages markedly better than others ?

And finally, is this the best place for discussing this API or is another list more suitable ?

I'm building an application to help with language learning and I am very excited about the API and the results that I am getting so far - keep up the good work !

thanks

--Rob Jones

[gsh...@chromium.org]

How much of the W3C api spec is currently implemented ? (I realize that the spec is not totally finalized)

The speech recognition portion of the W3C spec is implemented in Chrome. SpeechGrammar is not currently implemented.

http://dvcs.w3.org/hg/speech-api/raw-file/tip/speechapi.html

https://dvcs.w3.org/hg/speech-api/raw-file/tip/speechapi-errata.html

 

Is there any difference in functionality between regular Chrome and Chrome Canary?

Currently no significant differences between Stable and Canary in the speech recognition functionality.

 

Is there any way to get info on the state of a current connection between the client and Google speech recognition server, other than interim results?

Most of the time this works fine, occasionally I get a 'network error' and then some times it doesn't appear to do anything (even with seemingly clear speech) - I don't have a way to know where that problem lies.

I'm thinking of a log message something like '57 bytes sent to server'...

'network error' is the intended mechanism for reporting this to JavaScript, and there's no inherent way in JavaScript to determine where the problem lies.

If the problem recurs, and you can supply more information, that would be helpful.

  

Are there any 'accuracy' figures for recognition in each language ? I realize that is difficult to define - how well does speech recognition work in terms of % correct words for a given speaker and % correct for a given text spoken by a sample population, etc... are some languages markedly better than others ?

Yes, accuracy varies based on numerous factors, including language. Sorry I don't have specific numbers to share.

 

And finally, is this the best place for discussing this API or is another list more suitable ?

For Chrome-specific discussions of this API, this is a suitable list (Please add "Web Speech API" to the subject line).

For general discussion of of the W3C API, please use: public-s...@w3.org and http://www.w3.org/community/speech-api/

These links and more can be found in the tutorial at: http://updates.html5rocks.com/2013/01/Voice-Driven-Web-Apps-Introduction-to-the-Web-Speech-API

[P H]

Hi,

I'd like to work with the Google Speech API but I need to work with grammars. I read that you said on May, 30th that SpeechGrammar is not currently implemented.

So, I assume that it is not implemented yet, neither. I would like to know until when the implementation is planned?

Thanks a lot!

[P H]

Hi,

I'd like to work with the Google Speech API but I need to work with grammars. I read that you said on May, 30th that SpeechGrammar is not currently implemented.

So, I assume that it is not implemented yet, neither. I would like to know until when the implementation is planned?

Thanks a lot!

[gsh...@chromium.org]

Yes, it is not currently implemented. I cannot discuss future plans at this time.

[Andrew Prevost]

I have a more basic question about how the API has been implemented in Chrome.

I assume, but would like confirmation, that the Chrome implementation of the Web Speech API uses server-side recognition? That speech is merely being collected on the client side (i.e., on the machine where Chrome is running) and is then sent over the web to a Google speech recognition server somewhere where the actual recognition takes place, at which points the results are sent back to Chrome running on the client? Meaning the recognition would not work properly if you're running locally on a machine that is not connected to the internet, or is behind a firewall that blocks communication with all but a specific set of allowed sites?

Or is the actual speech recognition engine itself now packaged in to Chrome, meaning all recognition takes place locally on the client machine where Chrome is running, and pages that use recognition can be run locally even if no connection to the internet exists?

Thanks,
Andrew

[Tommy Widenflycht (ᛏᚮᛘᛘᚤ)]

Chrome is using server-side processing.

--
You received this message because you are subscribed to the Google Groups "Chromium HTML5" group.
To unsubscribe from this group and stop receiving emails from it, send an email to chromium-html...@chromium.org.
To post to this group, send email to chromiu...@chromium.org.

Visit this group at http://groups.google.com/a/chromium.org/group/chromium-html5/.

For more options, visit https://groups.google.com/a/chromium.org/groups/opt_out.
 
 

[Nagendra Kumar Goel (नगेन्द्र गोयल)]

Server side processing is fine. However when the API becomes part of HTML5, Let's hope that the standard is open so that any server could be used.

[Tommy Widenflycht (ᛏᚮᛘᛘᚤ)]

Right now the spec only specifies the JavaScript API but standardizing the client<->server traffic could be interesting as well.

[Nagendra Kumar Goel (नगेन्द्र गोयल)]

Without that It's a google project and should not be part of HTML5

[Peter Beverloo]

It's not part of "HTML5", and the Web Speech API document is not on the standardization track.  I suggest you raise your concerns to the W3C Speech API community group:

http://www.w3.org/community/speech-api/

Peter

[gsh...@chromium.org]

The Web Speech API specification is designed to allow for both server-side and client-side implementations.  A browser can implement it using any speech recognition engine.

The specification is designed to provide interoperability at the JavaScript API level.

https://dvcs.w3.org/hg/speech-api/raw-file/tip/speechapi.html#introduction

[Vladimir Vivien]

Hi,

Does anyone know when web speech will be on chrome for Android ?

Is that part of the future plan ? Will Chrome mobile have same support as desktop chrome for speech rec?

Thanks.

On Wednesday, May 29, 2013 4:35:20 PM UTC-4, Robert Jones wrote:

[PhistucK]

The Web Speech API is behind a flag in Chrome 30 (according to the blog post - http://blog.chromium.org/2013/08/chrome-30-beta-richer-web-on-android.html), the current stable release.

It is turned on by default in Chrome 31 (according to the blog post - http://blog.chromium.org/2013/10/chrome-31-beta-android-application.html), currently in beta, and it is generally a few weeks away from a stable release.

☆PhistucK