Asked
— Edited
Voice recognition is one of the most used controls within ARC however many struggle with accuracy and confidence, more so with the Pandora Bot control. So here are a few tips to improve the accuracy and confidence levels.
Improve Speech Recognition Start Speech Recognition
- Go to the Start button.
- Click on Control Panel.
- Click on Search box in the upper right-hand corner.
- Type "Speech".
- Click on Speech Recognition.
- Click on Start Speech Recognition.
- Click on Next.
- Select the type of microphone you are using and click on Next.
- Check the position of your microphone and click on Next.
- Read the sentence provided and click on Next. Note: You can proceed to the next page only after reading the sentences.
- Click on Next.
- Choose "Disable document review" radio button and click on Next.
- Choose "Voice activation mode" radio button and click on Next.
- Click on Next.
- Click on Next. Note: you can access the list of commands the computer can respond to later if you do not want to print it out at this time.
- Click on Next. Note: you can uncheck "Run Speech Recognition at startup" box if you do not want Speech Recognition to start automatically.
- Click on Start Tutorial if you are new to Speech Recognition or Skip Tutorial Note: you can take the tutorial at another time.
Accuracy
- Right-click on Speech Recognition Bar.
- Move mouse over Configuration.
- Click on Improve Voice Recognition from the sub-menu.
- Follow the Speech Recognition Voice Training tutorial.
To Train a word or Phrase
- Say "Open Speech Dictionary" OR
- Right-click on Speech Recognition Bar.
- Click on Open the Speech Dictionary.
- Click on Add a new word (You can also say it).
- Type in the word or phrase you want to train.
- Click on Next.
- Check the Record a pronunciation upon Finish box.
- Make sure speech recognition bar says "Listening".
- Click the Finish button.
- Click the Record button. (Note: Make sure you are satisfied with your recording before moving further. If not, repeat step 8.)
- Uncheck I want to make more modifications to the Speech Dictionary upon Finish box.
- Click on Finish.
When I have time I'll add some supporting images to this to make it easier.
does speech Recognition need to be on while using ARC
No, but training it is easier with it open since all of the options are there for you. Close it once you are done training.
Thanks Rich, this was much needed. I know it was for me. I didnt know some of this stuff. Do you know what the confidence box is within ARC's Speech Rec control is and what it does when you change the setting that is in there?
Thanks again, Dave Schulpius
I can only make an assumption but I would assume confidence is basically how good a match the phrase the computer heard was to a command. If you lower it, then it will accept a command as being good even if it's miles out... i.e. if you said "robot burn Geoff" it would probably hear it and accept it as "robot turn left". But if you were to increase it to a higher value it would not accept it as a valid command. Higher is very useful if you have commands like "robot turn off the lights" and "robot turn on the lights", or even "robot turn left" and "robot burn Geoff" (if you know someone called Geoff who you don't much like).
The adding a command/phrase to the dictionary is a bit of a hidden option, but it really helps improve accuracy/confidence, especially for the Pandora Bot control.
rich, your assumption is correct about the use of the confidence setting (part of my job is working with speech technology. Way more complex and expensive than sapi, and way more tuning than just confidence, but that is where we start).
OK, I've been wondering about that for a long time. I did I just did a few times and noticed a difference in speech recognition accuracy. Thanks for the confirmation an explanation.
I picked up a copy of Dragon NaturallySpeaking today and I have to say that the £75 I have ever spent. In fact I am typing this by speaking now. Dictation mode is really accurate I only had to correct one word in this entire paragraph.
I would highly recommend it. However I have yet to try it with ARC.
Looking forward to how it works with EZ-B builder and bow you set up windows/EZ-B builder to make it work. What mind of microphone are you useing? It would be a great help to see the different results you get with different microphones.
Currently I have a bluetooth headset connected, which is probably what I will be keeping now. It's not great, in fact it's pretty poor to be honest, I don't think it's any better than the kinect mic array I was using.
I think it'll just work with ARC but not entirely sure since the last time I used Dragon was when it was first released probably over a decade ago, it may just be coincidental but my other voice software I use seems to be working much better now.
Yes, the last time I used Dragon was about 10 years ago also. It was a cool little toy at the time and was neat to talk to you computer and see it type what you say. Most of the time anyway. I remember it also opened windows programs if you had Win 95 at the time. Anyway, I gave up on it as it was a little clunky and at that time typing was easier then having to go back and correct missed words on those slow computers back then.
I've very interested in how you will get ARC to integrate with Dragon through Windows. Please keep us informed on all the hows and whys. If you can get it to work with ARC I'm going to pick up a copy.
What version did you acquire? I see the Home version is up to 12.0 and they offer a premium version that does more.
It looks like I was wrong to assume it takes over from Windows VR. You can, however program it to control pretty much anything (somehow) but integrating with ARC may not be as simple as I had hoped. I know Tony (@toymaker) uses Dragon, he may offer some advice here...
Also, it still is a little clunky, or it is on my PC (which desperately needs upgrading!). I'll put more in to it when I eventually upgrade.
On another note, I did find a cool little program that saves voice profiles. Microsoft Speech Recognition Profile Manager Tool - I haven't tried it yet but it looks like it could prove to be very useful.
Rich, I currently use DNS11 with our own Ai and robot control software, but with EZ-B I use the Windows speech recognition engine as it been integrated into ARC. I am not sure if it would possible to use DNS with ARC as they are two totally different systems.
In my experience, Dragon is miles better than the Windows engine, but you have to buy it. I will in the future be looking into ways of trying to do connect DNS and also our face recognition (and eye tracking) software with ARC via something like telnet.
That's what I suspected once I was using it but the confirmation is always good to have (even if I did want to be wrong). I had, somewhat naively assumed it would just take over from Windows speech recognition.
I'll be looking at figuring out how to use it with ARC eventually but for the time being I have a lot of work to do moving all voice commands for non EZ-B related stuff to DNS... which should prove to be rather "fun"
Telnet seems to be the easiest way to connect/control ARC since it can use EZ-Script and is very simple. A ControlCommand to set something going or stop something, even passing variables through (which could be very good for the Pandora Bot control I know a lot of people want but can't achieve high enough accuracy).
For now though, stick to basic windows voice recognition (or what I've come to think of as voice activation since it's not very good at recognition without a set phrase list).