Author Topic: Where can I find Sphinx and Motion interface code ?  (Read 7709 times)

archived

  • Hello, I'm new here
  • Posts: 0
    • View Profile
Where can I find Sphinx and Motion interface code ?
« on: December 01, 2004, 11:51:36 am »
Hi,

I know I'm bit early, but would like to find code for interfacing Motion and Sphinx - I already use them standalone...

I've traversed svn site, but couldn't find them.

Any pointer where to search ?

Regards,

Rob.

archived

  • Hello, I'm new here
  • Posts: 0
    • View Profile
Where can I find Sphinx and Motion interface code ?
« Reply #1 on: December 02, 2004, 12:38:32 pm »
Hey Rob,

I put the old series 1 code up on the site for Motion and Sphinx.  We haven't touched them at all yet.  Under series 2 DCE is quite a bit different, although the concepts are the same.  Most of the old classes that had the name OC in front, are now in the DCE namespace without the OC.  The old OCMessage is comparable to the new DCE::Message.  The easiest way to get them to work is to 1) get the system setup on your local machine with the pluto admin web site, 2) go into pluto admin, device template, and create new devices for sphinx and motion, giving them the same data, commands, and events as the old one, then 3) run DCEGenerator to build a new Series 2 DCE Device, lastly 4) copy the command implementations from the old series 1 device into the new command stubs in the new series 2 device.

Motion worked fine for us in the old series 1 systems.  It crashed every once in a while, however the framework with the startup scripts just relaunches it everytime.  It never created a problem since all the DCE Devices are socket driven you can stop and restart them without hurting anything.

Sphinx was a huge challenge for us.  We got it working and it recognized voice commands over a microphone (lights on, off, etc.).  We also did an interface with Asterisk, so if you called in, the voice prompt would say "Who would you like to speak with?"  And it would transfer the call to the person accordingly.  The problem was the accuracy.  BEST case, we were able to get around 80%.  That made it interesting for techies, but useless for a consumer product where we would need to be close to 100% to keep from angering the consumer.  We hired a voice-recognition company to come to our offices for several weeks and work with us on it.  We got lots of different microphones, and they made a lot of tweaks--that's how it got to 80%.  But, in the end, they said that Sphinx just wasn't advanced enough to get much better than that.  They recommended the commercial product from Nuance--who does all the speech recognition for 411 information, the big airlines, etc.  Nuance's product worked well, but they required $1 million up front, and a guaranteed commitment of $5 million.  Since our volumes are pretty small, and the speech recognition is just 1 tiny part of the whole product, we ruled out that possibility.

The files are:
http://plutohome.com/download/OCMotion.tar.gz
http://plutohome.com/download/sphinx.tar.gz
http://plutohome.com/download/sphinx_support_files.tar.gz

The support files is huge--it contains tons of sound files and other files sphinx used.  The sphinx.tar.gz also includes the modifications we made to Asterisk to get it working over the phone system.  Neither of these were released, and we haven't looked at them in almost 2 years.

Hope this helps.

archived

  • Hello, I'm new here
  • Posts: 0
    • View Profile
Where can I find Sphinx and Motion interface code ?
« Reply #2 on: December 02, 2004, 09:34:32 pm »
Hi,

thanks for the code. I'll take a look and see if I can do any better.

Afterall, I'm speech recognition researcher...

BTW, is Sphinx2 you interfaced with ?

Will inform on any progress...

Regards,

Rob.

archived

  • Hello, I'm new here
  • Posts: 0
    • View Profile
Voicebox technologies
« Reply #3 on: April 12, 2005, 06:49:41 am »
Have you considered VoiceBox technologies for speech recognition...I have seen some interesting articles and talked to one in the group using it.  They only deal with OEM...not with public--might be a good marriage.  http://www.voicebox.com/

archived

  • Hello, I'm new here
  • Posts: 0
    • View Profile
Where can I find Sphinx and Motion interface code ?
« Reply #4 on: April 12, 2005, 09:39:39 am »
I just sent them an email

archived

  • Hello, I'm new here
  • Posts: 0
    • View Profile
VR
« Reply #5 on: August 11, 2005, 08:04:04 pm »
What ever happened with this?

As a current HomeSeer user (I just upgraded to HomeSeer 2 before finding out about Pluto, too, $@#!%^) I'm most interested in the VR.  I have a number of PZ-11 microphones and a Gentner AP800 that were slated for an HS integration.  If I switch over to Pluto (there are soooo many reasons why I'd do that), I want to be able to continue with my VR plans -- which are fairly complicated.

In a related note, has anyone gotten AT&T Natural Voices to work on Linux now that they've stopped distributing it for that platform?

Dantelope

archived

  • Hello, I'm new here
  • Posts: 0
    • View Profile
Where can I find Sphinx and Motion interface code ?
« Reply #6 on: August 11, 2005, 08:40:40 pm »
I talked to VoiceBox technologies.  They're interested only if we can guarantee them a good number of paying customers.  We don't have a lot of open source users yet, and we haven't released it for commercial sales yet, so at the moment, we can't guarantee them the volume they need.

I heard Sphynx (free, open source) has made major improvements too, but we've all been pretty swamped and haven't had time to look into it.  What does homeseer use for VR?  Does it work well?

archived

  • Hello, I'm new here
  • Posts: 0
    • View Profile
HomeSeer 2 and VR
« Reply #7 on: August 12, 2005, 01:15:58 am »
Quote from: "aaron.b"
I talked to VoiceBox technologies.  They're interested only if we can guarantee them a good number of paying customers.  We don't have a lot of open source users yet, and we haven't released it for commercial sales yet, so at the moment, we can't guarantee them the volume they need.

I heard Sphynx (free, open source) has made major improvements too, but we've all been pretty swamped and haven't had time to look into it.  What does homeseer use for VR?  Does it work well?


I'd steer away from the commercial VR products because it impedes adoption of an already difficult-to-implement technology set.  It can cost thousands to implement a decent VR system (less if you steer away from open-air mics)!

HS uses SAPI 5 to access Microsoft's own VR engine (L&H-based, I believe), which is speaker independent (but can be made more accurate with trained profiles) and, all things considered, very good.  A couple of HS users whom I trust are boasting 90% or better accuracy with their open-air mic systems after a great deal of testing and tinkering (VR has to come with a big dose of acceptance as to its limitations before you can use it without frustration).  It integrates with the AT&T Natural Voices (or NeoVoices) for truly ear-boggling continuous speech synthesis (really, really, really good).

The new version, HomeSeer 2, also uses a remoting application (.Net) called "Speaker" to push and receive voice and sound.  By speaking into the mic on one machine, you can send the voice (and a tag indicating its location) to the server, where it is recognized, processed, and the response directed back (or wherever the script says to send it).  That's nice in that it can eliminate a number of wiring considerations -- for example, those who want to set up VR using ONLY the computers in their rooms won't need to worry about open-air mics, mixers, and the like.  That said, it produces some weirdness because for those of us with more complex demands, we start wondering how to put 7 sound cards in a single box to keep the cost of multi-room VR down!  :D

Ideally, you'd like to see Pluto allowing voice tags to be associated with devices, events, and external applications.  You'd want to be able to define a grammar (or ship with a predefined, extendable one like HS) so you can easily say "Turn on the living room lights" -- or better yet, since it knows you are IN the living room based on the BT dongle or other ID mechanism -- "turn on the lights" and have it work seamlessly.

You'd also want to ensure that you can also use the microphones to talk to 1 or more other locations, thus enabling intercoms and room monitors and the like.

Thoughts?

Dantelope

archived

  • Hello, I'm new here
  • Posts: 0
    • View Profile
Where can I find Sphinx and Motion interface code ?
« Reply #8 on: August 12, 2005, 12:25:07 pm »
The framework is all there.  We have a very modular platform that allows any new devices to be added that can communicate with and control any other device.  When we have some free resources we'll look into adding VR.  If you know anyone on the Sphynx team you may want to see if they're interested.  Our DCE class generator means it only takes 5 minutes to have a working Pluto device, ready to compile and run, complete with stubs for implementing all the commands.

So 99% of the work is in figuring out how Sphynx works and how to make it do what we want.  Therefore someone on that team who knew Sphynx already could probably do it a lot faster, and we'd be happy to work with them--we can even give them a shell DCE device all ready to go.  Otherwise, when we have time to get to it it requires a bigger committment because of the learning curve for Sphynx.