PRO
jp15sil24
Germany
Asked
Resolved by Synthiam Support!
I am using Bing Speech Recognition with my new microphone, and it works well for recognizing phrases. However, I am encountering an issue with the wake word functionality. I have tried several wake words, such as "robot" and "computer," but they do not initiate recording. I am using the latest version of Bing Speech, which does not include the Voice Activity Detection (VAD) option. To troubleshoot the problem, I have only the Bing Speech skill loaded in ARC, with no other configurations. Did I overlook anything in the setup?
Related Hardware JD Humanoid
Related Control
Bing Speech Recognition
It sounds like you're experiencing an issue with the wake word functionality in the Bing Speech Recognition skill. Let's go through some steps to troubleshoot and ensure everything is set up correctly:
Check Microphone Configuration:
Wake Word Configuration:
Confidence Level:
Wake Word Sound:
Environment:
Testing:
Update and Restart:
If you've gone through these steps and the issue persists, consider testing the microphone with other applications to ensure it's functioning correctly. Additionally, you can try using a different microphone to rule out any hardware issues.
Feel free to ask if you have any more questions or need further assistance!
I am a customer support GPT-powered AI and still learning. If you have doubts about my response, please use the support documentation to verify my answer.
I tried other wake words and played with wake word confidence but same problem. the wake word is recognized, I can see the output but does not start the recording. At the moment, I use a switch push/release on d10, to trigger the start/stop listening
The log window will present information for diagnosis. See this image for examples...
As you can see, the word was spoken too softly at one time. Later, it wasn't clear and was below the confidence level. And the third time, it got it. This is a demonstration, but using a correct microphone in a quiet environment away from motors and servos will give you a good experience.
I tried several wake words, easy to pronunce like start,bing,robot but no ones trigger the start recording. I played with the wake word confidence from 0.50 till 0.95 . I unchecked the "play Wake Word sound with ControlCommand() as I remember it was an issue with a previous version of Bing Speech together with ControlCommand() but same thing. The start recording works well , my microphone is ok, I can say any phrase, they all are recognized, the phrases included in the bing speech , motor stop, motor turn left etc... are recognized as well.
update, after many tries and hours ,i could get one wake word working ok 2 times in a row !!. cant understand why it is difficult to get the wake word worked althought all phrases, recognized ones are working well. strange
Like support said, check the log as you use it. It'll tell you if your confidence is too low, you're speaking too quietly, etc. It's not easy to provide additional info without knowing any of the log messages. But I can say from experience that in most cases, it's noise near the microphone. If your microphone is on a laptop, ensure it's not near a fan or covered by anything. Often, a laptop mic can be on the front, which is a still place because it gets covered by your hands or your clothes if it's on your lap. It's not easy to give advice without knowing anything about your configuratio,r what kind of mi,c or what the messages are.
thanks for your tips. I was already thinking to install a remote switch at JD! Kind of walkie-talkie. Actually, I wonder if the best solution wouldn't be to use a remote switch, a push to talk. Well, just an idea.
Oh yeah - that's a popular approach. I find that approach to be better than a wakeword. Putting a switch on a microphone that starts recording when you press, and when you let go, it stops recording. When you see robots like Sophia and stuff on the internet, that's how they work. You always see a person holding a microphone and talking to the robot. They are pressing a button and holding it down when they speak. It's precisely like a walkie-talkie.
Otherwise, consider the awkwardness of a wake word. If you are trying to have a conversation with the robot, you have to say the wake word every time you want to say something. Using AI, that conversation would be very awkward.
Lastly, consider that if you remove the wake word and have the robot listen consistently, then the robot will listen to absolutely everything being said. If someone else is talking, the robot also listens to them.
The button is the best approach, IMO.