Slashdot Mirror
Princeton Students Develop Open Source Voice Control Platform For Any Device
rjmarvin (3001897) writes "Two Princeton computer science students have created an open source platform for developing voice-controlled applications that are always on. Created by Shubhro Saha and Charlie Marsh, Jasper runs on the Raspberry Pi under Raspbian, using a collection of open source libraries to make up a development platform for building voice-controlled applications. Marsh and Saha demonstrate Jasper's capability to perform Internet searches, update social media, and control music players such as Spotify.
You need a few easily obtainable bits of hardware (a USB microphone, wifi dongle or ethernet, and speakers). The whole thing is powered by CMU Sphinx (which /. covered the open sourcing of back in 2000). Jasper provides Python modules (under the MIT license) for recognizing phrases and taking action, or speaking when events occur. There doesn't seem to be anything tying it to the Raspberry Pi either, so you could likely run it on an HTPC for always-on voice control of your media center.
They need to develope a Raspberry PI with at least 4 usb hubs, I use them for all sorts of data collection applications like RF tag readers and security cameras.
When will there be new ideas? I've been in systems engineering since the early '80s, and no new ideas have appeared since the mid-'90s, except miniaturisation techniques. Computing has become dull.
There are tens of these sort of projects and this one won't run on 'any' hardware as it uses some heavy libraries that certainly aren't going to work on ANY of the embedded systems I use.
Well, I'm shocked at ther deceptiveness of the article! Shocked, I tell you, shocked AND appaled.
I mean there are *so* many devices these things won't run on. It won't run on my little PIC12F678 with it's 64B of RAM and 1K flash. It won't run on my typewriter OR daisywheel and it most certainly won't work on my wax cylinder player.
A clue: "any" in this context means there are no device specific restrictions limiting it artificially. But you already knew that.
SJW n. One who posts facts.
With a RaspberryPi you don't say? Quick, get a patent on this innovative technology that would be so mundane if it were implemented on a desktop machine running Debian or something.
I am becoming gerund, destroyer of verbs.
Are happening on a higher level now.
It won't run on my little PIC12F678 with it's 64B of RAM and 1K flash.
Nevertheless it would be nice to know approximately where the lower boundary is.
Saying that it runs on Raspberry Pi isn't really telling considering that the Pi pretty much has the specification of a full blown computer.
Even if it won't work on any 8-bit controller it would be nice to know if there is any possibility to make it run on a 32-bit controller without having to add external memory and storage.
The segment of 32-bit controllers in the 100MHz-range with around 64k internal ram and 200k flash is pretty big.
If one of those can be used it can be added to all sorts of neat stuff for a low cost.
If you have to add external memory you are moving to a segment of controllers with a different target audience and the neat motor control signals you wanted to use might not be available then.
Saying that it works on Pi is pretty much like saying that it works on a standard desktop without a fat graphics card. The Pi isn't very limited.
I already have this. I say "[son's name] - go do this for me" and he goes and does it.
There is a point worth remembering, If its on a network it's hackable. Why would you want a hackable always on microphone in your house. Sorry, there is no way in the universe i will have ANY live mic or camera permanently on.
A project that needs a full computer OS along with a high speed internet connection isn't really impressive. Now if he did it on an Arduino, that would show some progress.
"Format see colon. Why. Enter."
The fun things students come up with!
My son currently is in an engineering Graduate program in MA. He used to think they could try making cool things and maybe actually build and sell them.
Unfortunately most, if not all, can't take it beyond the classroom or home made use only. There is a huge list of patent trolls waiting on you if do.
Start here: https://www.google.com/?tbm=pts#q=voice+control&tbm=pts
He found this out the hard way :(
No, I don't magically know what they mean given no context. Thats the point. It won't run on any device, hell, it won't even run on any RaspberryPi since ... some of the licenses for those libraries themselves are potentially conflicting.
I'm not sure what magical fairy world 'any' device belongs to, but not a single one I can think of applies here.
But hey, why let reality cloud your inner fanboy, eh?
Persistent Volume manager for Kubernetes - https://github.com/dwimsey/openshift-pvmanager
Probably depends on what you mean by "network". As far as mere voice commands are concerned, some bit-banged unidirectional interface could easily send low bitrate signals into whatever network you have at home. How would you "hack" something like that?
Ezekiel 23:20
If the device can't run full-blown Linux with either ALSA or OSS support, then it sounds like you're out of luck. Full language models seem to be around 50MB of data, although I assume that simpler models could be used if recognition is constrained to a certain word set.
Compiled on my system, libsphinxbase.a is 298KB after being stripped, and the shared library is 302KB. That sounds like it's pretty far out of the size range that you're looking for.
It is pitch black. You are likely to be eaten by a grue.
_Of course_ 'sudo make me a sandwich' is in the libraries!
I find this very interesting. I was looking for an easy way of setting up always-on microphones with speech synthesis for intelligent home use.
I didn't plan on using a Pi though, but a few of the always-on full blown linux pc I have around.
Aziz, light!
I really like the way that these types of programs are taking us. It's about time that my computer starts listening to me while I'm yelling at it!
I've been using Blather myself, and really enjoy the results.
I do not want to detract from the great work they did to integrate and document these existing packages but I do need to point out that there is very little that is original about what they have done.
I have been using a very similar set-up to entertain my kids for several years and I just followed the instructions written by other people. The difference was it was not in one place and I had to research the options then stitch it together with a bit of scripting.
After saying the trigger word, you have to pause .. that's a bit ridiculous and annoying .. I doubt this would catch on .. for it to catch on, it needs to allow you to say a continuous sentence without pausing. The latest chip from audience.com has this feature (called VoiceQ). Their chip is for phones, so it should be possible to implement the same technology in software on a desktop CPU.
Wow Really very useful information.
Thanks a lot for sharing it with us. I will look forward to read more from you.
Could I share this information on my site.
http://webdesigningcompanyinchennai.in Web Designing in Chennai
jasper depends on the "CMU-Cambridge Statistical Language Modeling Toolkit V2" which is released under the condition that it will only be used for research purposes. Therefore, their setup can't be used for non-research purposes. I doubt that setting up my own home-automation system counts as research...