Micro_speech application "timing out"

Has anyone else see a timing out behavior in the micro_speech example code? I have been leaving my model running and it has repeatedly stopped working after about 24 hours.

Symptoms:
-Device stops taking commands
-Built-in led (not the RGB led) blinks in a repeating pattern (4 long blinks, 4short blinks)

This behavior has been observed happening across all of the things I have tested:
-device powered by two different computers (USB port)
-two different devices (I have two nano ble sense boards)
-different versions of the micro_speech code (both the plain example code and versions in which I changed the model)
-reset of the device makes it work again, but behavior reappears after about 24 hrs

I did not observe this behavior with the hello_world example code. It is possible I didn’t wait long enough, but I am under the impression that it is not the general TFLiteMicro code that is doing this on my devices.

My current guess is it is something in the microphone library, but that is a wild guess. Anyway, just wondering if anyone had seen and fixed this as I might be overlooking something trivial and I don’t want to re-invent the wheel.

Thanks!

Following up on this, the bug hunt continues. I was re-inspired to look into this as I got the same behavior immediately (no 24 hr delay) while trying to upload a sketch to a Nano 33 BLE Sense from my Raspberry Pi 4 running the Arduino IDE.

Based on the information in this link, I am now working from the assumption that something in the code is causing the mbed OS to crash.

So the good news is I have a lead, the bad news is now I have another use case failure to figure out. I’ll post if I get anywhere, I think I have some docs to read…

Hello Stephan.
This is Tensorflow my friend. It will timeout if you are not being active. You need to move around and click a few things every few hours to keep the environment from going idle.

I hope this helps, let me know if it fixes it or not.

Hello @FrancisElhelou ,
Thank you for your reply! In this case, I do not think Tensorflow is the problem. I tested by running the hello_world example sketch from the Tensorflow arduino library on my device for an extended period and I am not seeing the problem with the OS crashing. I will keep doing some testing and see what I find out. :slight_smile:

1 Like