PRO
manishabh77
USA
Asked
I'm having issues running the Speech Recognition robot skill in ARC Pro (version 2026.04.16.01). A script I use (titled "JD - Are You a Bird?") returns errors such as "low confidence" even after I adjusted the confidence threshold in the skill settings. I did not have this issue before.
I'm running ARC Pro on my PC and the speech recognition is configured as a robot skill within my ARC project. Any ideas what might be causing the sudden low-confidence results?
Related Hardware (view all EZB hardware)
JD Humanoid
by EZ-Robot
JD humanoid robot kit - WiFi-enabled, 16 DOF with metal-gear servos; easy, fun, educational, available from the EZ-Robot online store.
Wi-Fi / USB
Servos
24
Camera
Audio
UART
3
I2C
ADC
8
Digital
24
Related Robot Skills (view all robot skills)
Speech Recognition
by Synthiam
Windows Speech Recognition skill: detect custom phrases via PC mic, trigger configurable scripts/actions with adjustable confidence.
Bing Speech Recognition
by Microsoft
Accurate Bing cloud speech-to-text for ARC: wake-word, programmable control, $BingSpeech output, Windows language support, headset compatible

FYI - Before Athena addresses your question, you should be running the latest version of ARC before posting any questions or concerns. Please upgrade; otherwise, any advice you receive will not apply to you because the copy of ARC is outdated by almost 2 months.
A sudden increase in low-confidence detections usually points to one of these changes rather than the ARC project itself:
What to check first
1) Confirm the Speech Recognition skill settings
Open the skill’s kebab menu in ARC and go to Settings. Check:
Even if the threshold is lowered, if ARC is listening to the wrong mic or the mic level changed, confidence can still drop.
2) Verify the correct microphone is still selected in Windows
ARC uses the Windows speech engine, so if Windows switched input devices after an update or reboot, speech confidence can change immediately.
Check:
If you have a headset, USB mic, Bluetooth mic, or webcam mic, make sure ARC is not accidentally using a different one.
3) Check for background noise or robot audio feedback
If JD is speaking near the microphone, Windows may hear the robot’s output as input and reduce recognition quality.
Try:
4) Re-test Windows Speech Recognition outside ARC
Because ARC relies on the Windows engine, test the mic and language directly in Windows Speech Recognition.
If Windows itself is struggling, ARC will also struggle.
5) Inspect the exact phrase being recognized
Sometimes the phrase is recognized, but with a lower confidence because of:
If you’re using a phrase like "Are you a bird?", try a simpler test phrase such as:
If those work more reliably, the issue is likely recognition complexity, not ARC.
Why it may seem sudden
Low confidence can appear "out of nowhere" after:
Useful reference
Speech Recognition skill manual:
https://synthiam.com/Support/Skills/Audio/Speech-Recognition?id=16118
Microphone/audio troubleshooting:
https://synthiam.com/Support/troubleshooting/camera-audio-microphone-issues
A quick diagnostic question
Can you tell me:
If you want, I can also help you build a simple JavaScript test script in ARC to print the recognized phrase and confidence value so you can see exactly what the engine is hearing.
I realized what the issue was. I have also downloaded bing speech recognition and I discovered it is not compatible with the normal speech recognition module, even when I keep the bing recognition software on hold.
Thanks Manisha
You will get much more detailed and helpful responses if you use more words to describe the situation that you're experiencing and the desired outcome. If you write longer messages and explain what you're trying to do and what the outcome is, you will get more useful responses.
That makes sense - Bing Speech Recognition and the regular Speech Recognition skill are separate skills in ARC, and they can interfere with your workflow if both are configured around the same mic/listening behavior.
If you’re using Bing Speech Recognition, it’s best to keep the setup consistent and only have one speech-recognition path active at a time for that action flow.
A few tips:
If you want, I can help you restructure the ARC project so the speech skill you want is isolated and reliable.
Helpful links:
Since you found the cause, I’d recommend testing again with only one speech skill enabled in the project and seeing if the confidence issue disappears.