Voice and LSL

I was just wondering if there has been any word on any voice features that can be used in LSL. If not, lets start a discussion on what we would like to see.

My thoughts were that if the new voice engine can do some kind of voice recognition and then pass that recognised speech as text, it would allow us as scripters to create voice activated scripts. Even if it is just in the vein of creating a llVoiceListen and using a voice_listen event (for example) we can then have avatars using our scripts and saying "Door Open", and our scripts pick that up, determine the speaker and activate a door opening function (as a simple example).

If LL could find a way to integrate voice functions like this it would be a massive advance for LSL I believe ... depending of course on the accuracy of the voice recognition software lol.

According to the blog voice servers are seperated from game servers to avoid ingame lag. That would mean there are minimal links between voice and game systems, and any kind of LSL voice implementation (let alone "voice recognition" would add to those.

whoever lived through the nightmares of last weeks outages and lags will for sure want to keep those links at a minimum, if not sack voice completely.

You might be able to attach a voice recognition system to your line-out or line-in in your OS, then hack the SL client and send information between the two. Once you get the client compiling, inserting some basic messaging is not that difficult.

Honestly, beyond that, and due to the complexity of voice recognition, I doubt you'll see that sort of thing any time soon.

I made the post primarily as a kind of wishlist. Also, adding voice recognition to SL Voice wouldn't necessarily be as difficult as it sounds. If the voice servers are seperated, then the voice recognition can handle there. The only link that would be required would be passing the recognised text.

And voice recognition technology isn't difficult in itself to implement anymore. There are a few open source and commercial voice recognition packs that are remarkably accurate.

As far as using line in and out and hacking/compiling a voice activated version of the client ... welll ... what would be the point of writing voice recognition scripts if yours is the only client that can handle it.

Personally i think if avatars could interact with scripts via voice it can also help alleviate some of the (misfounded) concerns about adding the voice feature, as it then provides a useful function apart from just talking to other people.

Proper voice recognition takes some serious computing time.
Doing it for 100s of players simultaneously with any kind of accuracy... mayhaps in 10 years.

Not to mention it should work for all accents... and you'd still need to allow for chat commands for those that are speach impaired or generally not using voice (like yours truly).

So a while yet before we can use voice to tell Rover to fetch? :}

*IF* we get some form of voice recognition, it'll be triggered client side. To keep things simple I'd expect simple pattern matching, where the user has to "train" the command, like voice dialling on most cell phones. I'd expect some kind of addition to gestures before any to LSL.

I just don't believe this sort of feature is likely to ever be seen in SL. The processing involved is just too intense at this point.

Also how would SL know when you were chatting or when you wanted to execute a command?

Would we have seperate command channels like we do in text chat?

As mentioned in an earlier post is there in fact a voice recognition program that can work for all the different accents in the world?

What sort of lag would this introduce into the voice command system if the other issues could be addressed?

All in all I think it more likely we all have voice controlled cookers in our homes before SL responds to voice in that way. Which lets face it is far more use anyway!

There are some voice recognition technologies that are really solid (never flawless)...but these higher quality apps cost money...and neither you nor LL wants to implement these yet.