Author Topic: Google speech recognition integration into Voxcommando (Read 10059 times)

krysn95 · « **on:** July 15, 2015, 10:24:10 AM »

Hi Guys,

So, one of my buggest dream is a voice control system, but in my language --> Hungarian.

Unfortuatelly the MS Speech Platform 11 does not include hungarian languag pack, but I have an other idea:

Is it possibel to use the Voxcommando with the Google speech recognition?

With this Chrome built in, I can select Hunagrian language dictation, so this tool can convert my commands to text-->

https://chrome.google.com/webstore/detail/voice-recognition/ikjmfindklfaonkodbnidahohdfbdhkn?hl=hu

Does anybody has any idea to integrate this input mode (so the Google speech recognition) into the voxcommando?

Thank you in advance for your kind answers and sorry for my terrible English!

jitterjames · « **Reply #1 on:** July 15, 2015, 12:29:27 PM »

It would require me to write a whole new program that I do not currently have the time to prioritize.

At the moment you can accomplish this same thing using Autovoice and Tasker. It basically does the same thing as what you are suggesting.

In both cases this is the process that must take place:

1 - record audio and send it to Googles servers
2 - wait for a text response from the Google servers
3 - Send the text to VoxCommando and hope that it exactly matches the phrasing for a command you have created in VoxCommando.

In either case it should work with any language that uses the same alphabet.

Being able to do this all on a PC instead of having to use an Android device might be convenient, but it would not ultimately change what happens in this process, or what is possible.

Perhaps you can approach the maker of this Chrome app or extension you refer to, to see if they could offer the option to automatically save recognized text to a local text file, or to send it to a web URL. Then using the Watcher plugin for VoxCommando it could analyze the text. This might be a bit faster than using Tasker, and if you have a really good microphone on your PC it might give better results. I personally have no experience with writing extensions or addons for Google Chrome and no time to learn how to do that at the moment.

krysn95 · « **Reply #2 on:** July 16, 2015, 04:23:28 AM »

Thanks for your kind and professional answer.

The Autovoice and tasker kist simple not enough trustable. Sometimes listenning but sometimes not, so this solution is not the best.

So, if the app--> https://chrome.google.com/webstore/detail/voice-recognition/ikjmfindklfaonkodbnidahohdfbdhkn?hl=hu
could save the recognized text in a local text file, or to send it to a web URL, than it could work the Voxcommand as great as it normally?

thx!

krysn95 · « **Reply #3 on:** July 16, 2015, 04:28:51 AM »

this description help?--> http://www.labnol.org/software/add-speech-recognition-to-website/19989/

krysn95 · « **Reply #4 on:** July 16, 2015, 04:58:27 AM »

I find in develoopers mode the recognized text in the app. Does anybody any idea, how can I say for the pc, that do something if you find here this or that word? (Maybe with Eventghost?) Sorry for the lame question...

marcusvdt · « **Reply #5 on:** July 16, 2015, 11:53:21 AM »

I did quick search on the internet since I'll work on this subject somewhere in the future so I can perhaps implement open dictation for portuguese using Google's API.

The only way I imagine to do that is to implement a bit of automation to this process:

1-Record audio
2-Send to Google servers
3-Grab the text and use it either as input for VC as emulation or send that text directly to whatever application I may want to have open dictation enabled.

The solutions I have found for it on this quick search are:

1-You can keep the wav files that you say for VC (through PC mic input or Voxwav) in a folder for later converting to flac format or you can record directly in flac format by using sox (only PC mic input allowed).
2-You can send the processed audio file via a python script to google servers
3-the same python script would then work on the string returned to your needs. For example, you could try a complex word comparison and matching process so the spoken phrases recognized by Google could be matched according to their best matches against the command phrases that you set in VC. Or you could just send the recognized text directly with no processing for a simpler implementation.

Notice however that it is less likely that the recognized text from Google will exactly match the phrases that you have in VC.

Also, I did not try this method above. I just guess it will work.

And unfortunatelly, although I think I'll do this in the future, this is the bottom of my priorities for now.

Good luck!

krysn95 · « **Reply #6 on:** July 17, 2015, 10:25:43 AM »

Thakn for your pro answer.

What do you think, which soultion could be the most responsive and best option?

marcusvdt · « **Reply #7 on:** July 17, 2015, 01:37:21 PM »

I'm talking about performing all the three mentioned steps and respective solutions in sequence, probably using the python plugin for scripting the tasks in sequence.
If you don't like the recognition from tasker/auto voice, but https://www.facebook.com/pages/HP-Fit/1607287676153964likes the recognition from Google in your PC, then the problem can be the quality or setting of the microphone in your phone. I am mentioning this because I guess Google's online recognition engine is the same either for the PC or a mobile device.

krysn95 · « **Reply #8 on:** October 14, 2016, 09:33:56 AM »

Hi guys!

I am still not give up to use the amazing voxcommando with my own language (Hunagrian)

Would it be possible to use the Google hunagian language pack in Voxcommando, in real time, with this?--> https://cloud.google.com/speech/?gclid=CjwKEAjwhILABRDwo8mlqt6ug38SJACNSq_k0dhsOfJepxemdiBrVU0fNAtlNZkfYSL9XMKHavnHohoC-nXw_wcB

Thank you so much every post!

krysn95 · « **Reply #9 on:** December 06, 2016, 08:01:41 AM »

Quote from: jitterjames on July 15, 2015, 12:29:27 PM

It would require me to write a whole new program that I do not currently have the time to prioritize.

Could you estimate what time do you need to integrate the Google Speech platfrom into Voxcommando instead of the Microsoft Speech Platform?

I am sure the it could be a huge steep, because the Google hva more than 80 languages, so it mean much more user/possibility.

I want something in my language what is as professional as the Voxcommando in english, since lng time, so please think about it. Of course, If you wish I am ready to pay for your time!

nime5ter · « **Reply #10 on:** December 07, 2016, 11:14:40 AM »

I can't speak for James, who is not in the office this week.

I can say that apart from the development costs (which would be significant--not a couple hundred, or even a thousand dollars), the other challenge is that Google's Speech API is cloud-based, and they charge an ongoing license fee for developers to use the service.

https://cloud.google.com/speech/pricing

This is not particularly feasible for VC when you consider that most VC users refuse to pay for even the very modest cost of a one-time user license. Who would pay for the monthly usage of a global user base? What happens when Google changes its policies, which they have done regularly with all of their other services, requiring constant development updates or leading to the possible cancellation of the service entirely?

Again, while I cannot and do not speak for James, my guess is that an entirely new program as sophisticated and versatile as VC built on the Google platform is neither cost-effective nor sustainable for a small, independent developer in today's market--even assuming one is comfortable using an entirely cloud-based speech solution.

At some point, James may have ideas for a smaller scale solution that is not exactly what you wish but that may still help. I'm sure he'll let you know if something like that evolves.

krysn95 · « **Reply #11 on:** December 07, 2016, 11:25:33 AM »

Thank you very much for your answer!
Your points are absolutly understandable4

Do you have any other idea how can I use the VC in my language (Hungarian)?

nime5ter · « **Reply #12 on:** December 07, 2016, 11:39:28 AM »

I'm sorry, at the moment I don't.

I would like you to know, however, that James and I have discussed your problem more than once. We do sympathize and understand your frustration. If ever we come up with an idea that is truly usable, we will certainly share it.

It's also helpful to know that if there are some costs involved in implementing something, you might support the effort.

krysn95 · « **Reply #13 on:** December 08, 2016, 06:15:15 AM »

thank you so much!!

krysn95 · « **Reply #14 on:** December 09, 2016, 12:29:38 PM »

just one more question...

It seems that this Chrome extension can use the Google voice recognition in real time:
https://chrome.google.com/webstore/detail/speech-recognition-anywhe/kdnnmhpmcakdilnofmllgcigkibjonof

Would it be possible to use this method in any way into VC?

thank you!

jitterjames · « **Reply #15 on:** December 09, 2016, 06:32:57 PM »

You could use this extension in a web browser and send the recognized text to VoxCommando (by watching the text field for changes) but it would only work as long as the web page remained open and focused so that's probably not really useful for anything more than testing.

I don't know of any way to access Google's speech API for free without either using Chrome or Android.

nime5ter · « **Reply #16 on:** December 05, 2018, 05:46:06 PM »

Quote from: krysn95 on July 15, 2015, 10:24:10 AM

Hi Guys,

So, one of my buggest dream is a voice control system, but in my language --> Hungarian.

Unfortuatelly the MS Speech Platform 11 does not include hungarian languag pack, but I have an other idea:

Is it possibel to use the Voxcommando with the Google speech recognition?

With this Chrome built in, I can select Hunagrian language dictation, so this tool can convert my commands to text-->

https://chrome.google.com/webstore/detail/voice-recognition/ikjmfindklfaonkodbnidahohdfbdhkn?hl=hu

Does anybody has any idea to integrate this input mode (so the Google speech recognition) into the voxcommando?

Thank you in advance for your kind answers and sorry for my terrible English!

Good news. James has added this feature to the TCP plugin in the new, beta version of VoxCommando (VC 2.2.4.5)

PegLegTV · « **Reply #17 on:** December 05, 2018, 07:41:49 PM »

Great work!

feels like an early Christmas gift

, I can't wait until I have time to test this out.

Kalle · « **Reply #18 on:** December 07, 2018, 05:47:13 AM »

It works fantastic

Have you ever tried to speak a language other than the one you chose or even a mix of languages? This works surprisingly well. I can also explain what advantage it has:
For example, if you are looking for or want to play a movie or music track that is different from your native language - in my case - use German language input with English or French music track.

The new plugin brings a lot of new possibilities - great job done James

jitterjames · « **Reply #19 on:** December 07, 2018, 04:51:04 PM »

I have noticed that when I am in a non-english language the Google SR engine still picks up some English words quite well. But I think English is detected when you are set to other languages and that's it. You can't set the recognizer to French and then speak German and expect it to understand the German.

I also noticed that some words don't pick up well. If you use a prefix like Jarvis it will work fine in English but not very well in other languages. Ok Google is a very good choice if you want to use the same prefix in multiple languages since Google seems to think this is a very important phrase in every language on Earth.

Brian · « **Reply #20 on:** December 31, 2018, 11:43:05 AM »

Wow, this is awesome! Thanks for this. I tested out the beta and works well.

This works great on my old webcam (open air) mic. I also do not use a prefix. I use my normal volume levels and it works probably 95% of the time.

I'm at work now and no access to Vox but Vox does freeze after a while. Not sure what the issue is and will post logs later. Both the tcp and vox freeze (perhaps after idle. I'll wake up and it's frozen) and have to end the task. I also use Kodi 18 RC3 and gets most of the commands but I can't get the tv guide to load. Without the tcp going, Vox will open the tv guide but with tcp and GS going it will not open even though Google heard it correctly. Also, "browse Tv shows" won't register but "browse televsion" will.

Thanks!

jitterjames · « **Reply #21 on:** December 31, 2018, 06:54:14 PM »

Hi Brian. Glad to hear you are enjoying the new toy.

Quote from: Brian on December 31, 2018, 11:43:05 AM

Also, "browse Tv shows" won't register but "browse televsion" will.

If a command does not register then it's most likely just a matter of spelling. Compare what Google is recognizing and how it's spelled to the spelling of your phrases. In some cases TV might have been spelled TeeVee or T.V. etc. in our default command sets because those sometimes seemed to work better with the Microsoft engine. I don't think that capitalization is important though.

Quote from: Brian on December 31, 2018, 11:43:05 AM

...but Vox does freeze after a while. Not sure what the issue is and will post logs later. Both the tcp and vox freeze (perhaps after idle. I'll wake up and it's frozen) and have to end the task. I also use Kodi 18 RC3 and gets most of the commands but I can't get the tv guide to load.

I'm happy to look at logs, but finding the cause of freezing can be almost impossible sometimes.

The other issue with Kodi is certainly very curious. Not sure what's going on there unless there is some sort of port conflict but I can't think why a port would be involved specifically in opening the TV guide. Does it involve a device like a PVR? In any case you'll probably need to look at both VC and Kodi logs to figure that one out.

Happy New Year.

sweefer · « **Reply #22 on:** June 16, 2019, 05:52:25 AM »

Does the Google also work in German? that would be great

Kalle · « **Reply #23 on:** June 16, 2019, 10:03:54 AM »

Yes it does - you can set the language within the TCP plugin (it would be a good idea to watch the video above)