Open Source Speech Recognition: simon

[grey0powerlessnesss]

Best wishes for the project. Staying with Win7-32 (Nuance's DNS), but Ubuntu-32 (Super Os) as second boot. Linux-32 has no 4gb memory limit, so no need for 64 bit IMHO.

USB digital microphone (Logitech) works better than any analog-digital conversion. Look for hardware noise-cancelling inbuilt into micrphone.

Medically Retired It Consultant, Australian Capital Territory

[basheersubei]

Wow! good work! I love the table function!

[AnushrutG]

Is there anything here like a command or something, like "make me a sandwich" ??

Please help!

[VideoProfess0r]

I have developed my own algorithm that is highly useful for video/audio editing and am looking for a partner to help me commercialize it. Contact me if interested.

[edjsage]

Any hope for a 64bit version?

[justbedahr]

@edjsage We already provide (and always did) provide 64-bit packages for a variety of Linux Distributions. You'll find appropriate packages on our sourceforge page.

We don't yet provide a 64-bit binary for windows as packaging for windows is really not straight-forward. Maybe for 0.4...

[edjsage]

@justbedahr I must not have looked hard enough. Thanks!

[theNewCodingFrontier]

I was going so say something along the lines that hlovdahl wrote..then I realized..RTFM...sigh..

[justbedahr]

@theNewCodingFrontier The introductory wizard (simon 0.3) should prompt you to install scenarios and set up a base model. If you do that you will have a fully functional simon in less than five minutes.

The handbook does of course also tell you that but it also contains a lot of information about the logic behind simon and how to create new scenarios. That's also why it is quite long :)

[theNewCodingFrontier]

@justbedahr oh, thanks alot I also realized what I was doing wrong (I wasn't loading models properly) also I had an issue install the RPM because QT4 is now 4.7 and the current simon 0.3 calls for QT4 < 4.7 so I had to edit it out. Just a heads up incase you're getting bug reports. Thanks again!

[justbedahr]

@theNewCodingFrontier Thanks. The issue with qt is that we were using QtMultimedia which was part of Qt - up until 4.7 when it was refactored to be part of QtMobility. Newer versions of simon (including the package of simon 0.3.0 for Ubuntu 10.10) use the QtMobility version and require Qt >= 4.7 as a result.

[theNewCodingFrontier]

@justbedahr Oh, I am using it for fedora, but I got everything working it's just simon isn't responding. I have trained several times with all my words (Using with JaNET) and all my samples give me the warning that I should re-record because background noise. Thing is I'm using a lap top so no matter the volume there will always be a little static (I have tried it at all ranges). I also don't use mic boost, what is the success rate with this type of mic? Thanks again

[justbedahr]

@theNewCodingFrontier Not that good, I am afraid. Even if you get the recognition to run you the recognition rate will be quite poor in our experience. This really is like competing in a race with a 500$ scrapyard car. By spending a tiny bit of money your recognition rate would be a _lot_ better. Try to find any old headset and it should work much, much better (you'd need to scrap your training data as well if you trained your model with the old mic).

[justbedahr]

@theNewCodingFrontier Your problem is most likely related to simons voice activity detection that segments the audio stream for the recognition. This tries to differentiate between voice and silence. If the background noise is too high, this segmentation fails and the recognition never gets anything to recognize on. Fourtunately this is configurable (Settings > Configure simon > Recordings > Voice Activity Detection). Try playing with the "Cutoff level" (increasing it) until it starts working.

[justbedahr]

@theNewCodingFrontier Btw. We have a support forum at sourceforge (tinyurl°com slash 5wxqsva). If you are still having problems we could maybe move the discussion there as it is imho much more suitable for this than this comment section...

[xyz98741]

LOL EPIC FAIL. You can't even articulate sentences or nothing.

What a farce. It's open source though, so you'd expect it to be nothing but a pile of garbage.

[theNewCodingFrontier]

@xyz98741 I would be surprised if you could even spit out a hello world program.

[justbedahr]

@xyz98741 You are right, full dictation is not yet supported. There is a lengthy explanation of why here: tinyurl°com slash kk9ec8

Other than that, we are always open for constructive criticism! You can also reach us directly per mail at support ate simon-listens ° org

[BitterToken]

1> next> 2 > 6> enter

OMG I just killed someone O,o

[Decide85]

it so not star trek.

[19bomba92]

great! ^^

[grThetrojan01gr]

Now, I am speechless!

Impressive is also the fact you've made it open source. Congratiulations! :)

Best wishes for the project.

[wlijiu]

you need to start a Simon server, named simond or as a GUI ksimond.

[Tverskoy1982]

Круто! Развивайте дальше.

Cool :) Good luck!

[BrainSeepsOut]

It would be more impressive if it could recognize something else than prerecorded commands.

[justbedahr]

It doesn't do mere pattern matching but actually "learns" your speech patterns. So you could add words that simon never heard you say before and simon would still be able to recognize it.

Now be impressed :P

[emomagica]

@BrainSeepsOut Yes it is pure bs. We need a true speech 2 text solution in linux and all I ever see are these garbage programs where you bark commands to open programs. I'm tired of linux I'm gonna get a mac.

[hlovdahl]

I would be even more impressive if it would be simpler to get it up and running. To me it just keeps telling:

"Couldn't start recognition because the system reports that the recognition is not ready.

Please check if you have defined a wordlist, an appropriate grammar and recorded a few trainings samples.

The system will then, upon synchronization, generate the model which will be used for the recognition"

[justbedahr]

Please read the manual for detailed instructions.

You might also be interested in our article on the KDE dot (search for simon kde voice) which talks about why simon is relatively hard to setup and why this is a good thing.

[schmidtbag]

wow very impressive, i'm going to download this now

[BladeYang]

Hello.

Currently I'm looking around for speech recognition software because I'd like to integrate such a thing in a multimedia project that involves it in order to teach a language. I'm looking for something along the lines of what Rosetta Stone uses that could be used in a sort of point and click adventure game. Think of the shows that are meant to teach kids foreign languages but in multimedia form. Know anything that can help me?

[justbedahr]

No sorry. But I am currently working on something similar (but much more basic of course) with a language school called ABC International. The finished software (if the concept works - we are still not quite sure :) will be released under the GPL too...

[BladeYang]

Wish you the best on that, if you can and if the concept works, leave me a message reply haha.

I'll look into using volume levels to detect different sounds for some specific words.

[Ktoso]

Looks great, hope it to keep improveing. Are you thinking about immplementing something like "five next" (vi'ish... :)) to speed up navigation like in the mail window where you had to say "next" quite a few times?

[justbedahr]

Nice idea... Maybe I should think about that :)

[VShaderPl]

Looks nice, but I can't image controlling my PC that way (too much speaking). I hope you'll improve it :)

[fusioniki]

Great! Looks like this is going to be awesome.

[justbedahr]

Use mail: grasch ate simon-listens°org

[justbedahr]

I am sorry I can't comment more than this

[paulopina21]

How did you get the English Dictionary?

I Installed into my Windows 7... But the words I had trained just doesn't work.

Do you used or knows any tutorial that I can use to get started with Simon?

Thanks!

[justbedahr]

You can import the English Voxforge dictionary

[cumulus0007]

I've spent a whole day on making Simon work, but I just couldn't. This software is awesome, but way too complicated for a beginner. I just want a wizard or something, not a whole manual to read! When I try to activate Julius, I get:

Couldn't start recognition because the system reports that the recognition is not ready.

Please check if you have defined a wordlist, an appropriate grammar and recorded a few trainings samples.

I'm desperate, please help me!

[justbedahr]

Hi!

Because of simons architecture it is not really easy to get started.

We are trying to improve that with the version 0.3 which is currently in development.

In the mean time, however, i can only really refer you to the manual which explains the concept and really all you need to know about simon - and is not _that_ long :)

Greetings,

Peter

[Denis150386]

Is it uses PortAudio or no?

[justbedahr]

Yes. Version 19.

[bluehash]

Pretty awesome. Can you call system commands via the terminal. This will really open possibilities like running scripts. Good work!

[justbedahr]

Yes of course. That is what executable commands are for. You can even specify the working directory if that should be needed.

[Meskarune]

This blows my mind.

[test123ok]

Does it give audio feedback ? I want my mom (who is blind) to be able to do something as simple as control a music player. ideally it would be great to say the name of the song and the software do the text search on the music software.

Another nice feature would be to read the headlines from a news source and the user should be able to select a particular headline and read the details.

[justbedahr]

Hello test123ok!

I am afraid this is not easily achievable...

simon is not (yet) suited for dictation (large vocabulary base) like it would be needed for searching for music (the non-standard names of artists and tracks makes this especially hard).

However, you could try this:

List all albums. The currently selected album is read out loud. The user can then navigate up/down. (selections are read out loud). Upon finding the wanted album, open it (with voice) and the titles are read out loud, etc

[justbedahr]

Hello again!

As for the news: There also are possibilities to limit the vocabulary. The software could read the headlines and associate them with numbers. She could then read the news by saying the associated number.

Both (news & audio) would be possible with current open source technology.

Contact me (grasch ate simon-listens ° org) if you are serious about implementing either or both of those projects.

[fax8]

this is really cool.. I'm going to test it now!