Any Success in speech recognition on Maemo?
 

hi everybody,
i have been search for a speech recognition software that works on maemo.
i have seen a lota topics about this but there was no solution or any sign that someone had succeed neither.

i wanna use speech recognition ability in an n810 controlled robot project.
it would be great to have such a feature.

regards.

Re: Any Success in speech recognition on Maemo?
 

Hi, Buddy, I've ever ported sphinx to N800, and of course it should be workable for Maemo 5.

Re: Any Success in speech recognition on Maemo?
 

Hello pawpawyoung,
do you think that is ever possible to use sphinx as a component for voice dialing in N900?

I see that this was taken in consideration in a brainstorming, but i don't see any follow up.
http://maemo.org/community/brainstor...oice_commands/

Re: Any Success in speech recognition on Maemo?
 

no actually i got an n810 too
can you share the information about how to install it?
pleaseeeeee pleasee im begging you

Re: Any Success in speech recognition on Maemo?
 

Aren't there many other threads on this topic already? We really need to get more focused.

Pocketsphinx works on n900 from http://0xab.com/?p=93

Need to enable repository from instructions there, and then you're in for some effort to make it work. Need to build your own UI, which isn't too hard, but time consuming. AFAIK, nobody's packaged up a nice GUI yet.

I'd really like to be able to plug pocketsphinx into emacs for dictation/command execution. You could send SMS and call contacts easily this way. http://sumoudou.org/%E7%9B%B8%E6%92%...U%20Emacs.html The easiest way I can think of doing this is using a language model that simulates a keyboard, but haven't kludged this up yet, nor seen anyone else who has.

Pocketsphinx also works well on n8xx from what I've seen. http://www.youtube.com/watch?v=OEUeJb6Pwt4 ...but this is still far from the usability of google's cloud based speech recognition. Nice that you don't have to send it all over the web though, and it's much more open/configurable if you have the time/effort.

Re: Any Success in speech recognition on Maemo?
 

Hello,

I've had a run with the repository on 0xab.com
Couldn't get things to work.
* Here's what i tried: sudo apt-get install chicken-qt
I got a problem with libqt4-maemo5-dev
* I tried to install the sphinx packages found here (http://0xab.com/n900/repository.html) under voice recognition.
I also had problem with dependencies...

Could anyone give us a step-by-step? Getting voice recognition up and running would be cool! Perhaps it's a problem with PR1.2?

I have worked on sphinx4 using Java, but all under windows.

Anyway any help would be appreciated.

Thanks,

Re: Any Success in speech recognition on Maemo?
 

I hacked this to N900. Works quite nice.

If I have more time maybe I will implement some basic voice dialing, shouldn't be that hard.

If anyone wants to try it, follow http://0xab.com/?p=93. Please note that you need to change alsasrc to pulsesrc.

Re: Any Success in speech recognition on Maemo?
 

kinipyon,

Great news! Could you tell us how to get this working on the n900, perhaps a small step by step? I have had trouble with dependencies and packages, and would like to experiment too :)

Any help would surely be appreciated.

Thanks in advance :)

Re: Any Success in speech recognition on Maemo?
 

Have you seen this voice activated browser:

http://www.youtube.com/watch?v=gjIUv84U_E0

Re: Any Success in speech recognition on Maemo?
 

http://code.google.com/p/voximp/
Download voximp.tar.gz to anywhere (MyDocs included). Unpack it (I did it in Filebox).

Install through apt-get

gstreamer0.10-pocketsphinx
pocketsphinx-hmm-en-hub4wsj
python-pocketsphinx

(I suppose it's bare minimum)

In voximp.py change alsasrc to pulsesrc

Run in X Terminal
python /path/to/voximp.py

It will ask you to copy/move three files (included in voximp.tar.gz you unpacked earlier; something like 9****.dic) to /home/user/.config/voximp/

When most problems are solved, running in X Terminal
python /path/to/voximp.py
will give you
letter X
command X
/bin/sh: xdotool not found

It means that it listens and recognises, but lacks the way to give it to operating system as pressed key of the keyboard.

I have taken xdotool of Debian for armel processor.
http://packages.debian.org/stable/x11/xdotool
Here it is (just the binary):
http://sharesend.com/info/uqr8g
Works-for-me.
Of course, you could compile it from source and publish to Maemo Extras; you are welcome! I'm lazy.

Now, running in X Terminal

python /path/to/voximp.py &

Will give you, besides long output in X Terminal,
ability to prints letters, and execute commands (cut, copy and paste, I don't know what else) in any program

Warning: ' &' at the end of line means that voximp will work when you close the X Terminal

When you try to use X Terminal (kill voximp, for instance) later (with voximp running), you will get spaces and random letters (depends on level of noise around you, actually).

Good luck! I'm going to test voximp later. It works but it's kind of unusual.

EDIT:
I still cannot open Fennec by voice command.
The problem is, it can recognise Fennec as partial (if it's pronounced aloud by espeak), but after several partials, the final command ISN'T Fennec. It can be T or some other letter.

I don't have ideal testing conditions, though. I have loud computer keyboard, sand in N900 speakers, and dust in N900 microphone.
So it's not surprising that sphinx rarely gives correct recognition.

The problem is, how do partials work and why the 'Fennec' partial was swallowed instead of being used as final command?

EDIT: I have read the .dic file. The pronunciation (mine or dic's) is awful. But at least sometimes it recognises (final) when I say Fennec. And I stopped outputting partial. And:
If you want to use voximp to enter text, you need to find somewhere full corpus of your language unless you are going to spell it letter-by-letter.

And with Fennec, I don't know what command could open a new tab in existing window.

Re: Any Success in speech recognition on Maemo?
 

Where did i screw up?

Re: Any Success in speech recognition on Maemo?
 

Ok, i didn't rename the config to .py

Re: Any Success in speech recognition on Maemo?
 

voximpconf.py must be in the same folder as voximp.py
Both are supplied in voximp.tar.gz

P.S. I have tried to make pocketsphinx recognise words spoken by espeak. No success. Most likely, it's caused by dust in microphone. What's the easiest way to record some audio from microphone? Some gstreamer line for X Terminal is welcome.
If microphone turns out to be ok, I will create my own acoustic model. Recognising words spoken by all people in the world is too imprecise; there are different accents, after all. And I prefer my N900 to recognise only my voice, and recognise it perfectly, than recognise voice of anybody in the room, and fail at it.

Yes, I have tried it at human speech, too. It gives a lot of vowels, and only rarely consonants. Well, maybe, speech corpus is too bad, I will look into it.

But right now I will need to check quality of microphone.

Re: Any Success in speech recognition on Maemo?
 

Could it be that the speech recognition engine is better at recognizing human speech than machine speech? I myself find simulated speech to be hard to understand way more often than real human speech...

Re: Any Success in speech recognition on Maemo?
 

very interesting, thank you guys!
to date voice recognition is present in every smartphone through free apps. therefore there should be an open and free way to have it on n900. we definitely need a package in extras-devel.