Home Robot Projects In ARC

Ok, so I have a question that may spark some to chime in here... Say some of you may have designed a robot that you use to show off at a party event. One of its task may be to roam about the room, offering up to serve/deliver drinks to the guest. How successful have you been with dealing with the party noise when it comes to giving the robot a verbal command to do something, or it being able to respond to a verbal command from a guest maybe? I myself am still dealing with any external noise that causes the bot to do something on its own. Any thoughts/comments from anyone?

For me and events I use this: https://synthiam.com/Support/Get-Started/Tutorials/speech-recognition-with-push-button-microphone

Push button to talk like a walkie talkie. It’s the only way I ever use speech recognition. Otherwise the robot has no clue if you’re talking to it or someone else etc.

I think trying to use chat gpt in audio mode is a good example of why the push to talk makes sense. Chat gpt in audio mode is difficult because it picks up everything and doesn’t know when you’ve stopped talking. So small pauses can trigger an end of a sentence.

Push to talk fixes all that

Like you and others I had trouble with false triggers due to background and crowd noise. I decided early on that the only way to overcome this issue was to be the only person my robot heard and responded to. I found a true high quality over the ear microphone, the Plantronics VOYAGER-5200-UC. It works directly with my robot's computer running ARC through a USB bluetooth dongle. It has amazing noise canceling and works great with Windows and ARC. One downside is that a good quality mic like this can be expensive. This one sells on Amazon for between $100 and $135 USD.

I then set up a "Wake Word" in ARC so the only thing the software listens for is that wake word before it starts listening for commands. Exactly like Alexa. Before Synthiam added the Wake Word feature to the Bing VR skill I had a separate VR skill set up that only listened for that wake word. After it heard that word its only job was to trigger a script that started Bing listening for commands. There's no need for that extra step now that the Bing SR skill has the wake word already built in. I was skeptical about trusting Bing's built in wake word but I was pleased to find it works perfectly.

This all works great for me. I get at least 99% successful sound recognition and the only time I get a false trigger is when it hears me say the wake word when I don't want it to respond. My word is "Robot". So if I'm just chatting, have the mic active and say "Robot", Bing will start listening for a command. I also have a visual light effect happen during the recognition period I have set in the skill (about 4 seconds), I also have a small movement animation on the robot so when it hears the wake word so I know when it's triggered. This added animation also helps cover up the inherent lag time it takes for Bing to send and receive what it hears before the command is able to start. This delay is just a few seconds but just standing there in silence before Bing responds back seems like forever and not natural. Having this animation helps cover up that delay.

I love @DJ's method of using a push button for total control. It's completely foolproof and works 100% of the time (unless you forget to push the button. LOL). It's also a lot less expensive than my $100 + over the ear mic. However, what I like about my method of using this high quality over the ear mic, with noise canceling, is that it's seamless and frees up my hands. I'm untethered and I don't have to remember to push a button. With either method, you'll still have that lag when the VR skill sends and receives your command.

I like both the suggestions. Some kind of mic sounds like the right direction to go. Right now I simply use the mic built into my laptop. I do already use a wake word, and have been experimenting with how quickly I have to say the command then pause the listening. It really is annoying when nearly any odd sound other than a tv can cause a false trigger. It could be a low hum from an appliance, a siren going by, even if I breathe, sets it off .

Interesting that random sounds get by the wake word. Maybe try a different, more unique wake word along with a higher quality microphone?

Yeah, it's crazy. Even tonight, while running more gripper testing, I'm like sitting right in front of my laptop trying to give a command. The tv volume is off, only my a/c running, and yet the Speech Req would still complain that there is too much noise. Next thing I know I see both robot arms flailing about trying to do a test pattern that I did not ask it to do.

I think the solution there is to increase the wake-word confidence threshold. By default, it's quite low if I recall. As I said, I don't really trust it, so I don't use it. even my Amazon Alexa picks up often with music or movies on, etc...

Just had another thought, and I may have asked this once before. Of all the AI skills that work with ARC, which one has been everyone's favorite? Easy to use, Easy to program? I've been using the PandoraBot skill mostly, Seems easy enough and is free Also have tried the Chatgpt, a little.