No detections for today?

[dbrianhendrix]

It appears this is becoming a regular occurrence -- the system is staying very behind. I'm showing NO detections for today and the log is showing it working on recordings from yesterday evening.

[dbrianhendrix]

[Svardsten53]

At least two of us here in Sweden also had this problem earlier today. Restarting services helped and issue seems to be solved, This issue has been discussed in other threads as well.

[dbrianhendrix]

I rebooted the Pi but I'm still not caught up. Currently, I'm still 4 hours behind but keeping an eye on it.

[nilspupils]

Had the same problem yesterday, resolved with rebooting.

[grajohnt]

Came here to post a similar problem - neither rebooting nor restarting services helped. It's now 6 hours behind, but verrry slowly catching up. (Note, running a 3B+ here, and just enabled the zram swap).

I'm still very much in 'setup' mode with mine, so I could either manually delete the files to accelerate the processing, or alternatively blitz all of the data and start from scratch. Any less-radical suggestions?

[grajohnt]

Watching the log, I think this is going to be a recurring problem.

After some hours of 'catching up', the timing of processing past audio chunks suddenly leaps from 3-4s for each to 30s each. Once this happens, the queue will fall further and further behind, given that they're 15s clips. Restarting services (or rebooting) fixes the problem, but it seems to happen again regardless. (I'll screenshot my log if I catch it again).

Any idea what might be causing this?

[ehpersonal38]

How many do you see when you issue this:

I see just 1 here!

[grajohnt]

This is what I see:

throttled=0x80008

Binary: 10000000000000001000
10000000000000001000
ISSUES DETECTED
Soft temperatue limit active
Soft temperature limit has occurred

[grajohnt]

...but it's running normally right now - I'll keep an eye on it

[grajohnt]

OK - scratch that theory...

It's running behind now, and not showing that it's throttled, and temp is at 58.5C, so no reason to do so.

Interesting, it's now stuck in a state where it's still recording audio, but not processing it.

Here's htop:

It looks like that ffmpeg job is what's holding things up?

[grajohnt]

OK - Final update here - I upgraded to a Pi4 (same SD card), and have zero problems now and don't need reboots.

I'd perhaps suggest that you recommend that you only use a Pi4 for this.

It's a great project, and much appreciated - thank you!

[nilspupils]

So it also happend to me, system was stuck with a "file not found" error looking for a wav. After deleting the last remaining wav file and rebooting twice, system is happily catching up on yesterday's files.

[GeekNJ]

I also ran into this for the first time after running my pi 4 for over 3 weeks. Here are my observations, but if there is any analysis that would be helpful, please let me know.

I noticed it was not displaying any new detections on the overview screen so I rebooted. I did not look at any system info before rebooting.

When it came back online after the first reboot, I did a View log inside the app and saw it was now processing recordings from the previous evening. It took quite a while to catch up. It looked like about 3 seconds for each 15 second audio so that is 12 seconds for every minute it hadn't processed or 12 minutes to process each hour of backlog on my pi 4.

After it finally caught up, I noticed that though the log seemed to indicate it was processing as normal, and there were still birds being recorded as I watched the live & spectogram pages, the analysis was showing 50% or less for 20 mins or more though I could hear unique birds just fine. So I rebooted again but needed to head out to dinner so I didn't stare at the pages. After the reboot post catch up, everything seemed to return to normal.

It could have been a coincidence but I usually have 5-25 detections in any given daylight hour and I could see then outside and hear in the recordings, but they just weren't hitting the confidence threshold. They were identified correctly but just a 50% or lower confidence.

One other observation is the Reboot button in Tools / System Controls doesn't seem to work for me. I click it, and it seems to prevent access to the app, but it never seems to fully reboot. If I run a ping to the IP address, it will just show a timeout. If I pull the power and restart, it always boots up in under 40 seconds. Next time I can also try to ssh in from my computer and do a sudo shutdown -r now to see how that works out.

[mcguirepr89]

Hi @jflyons

There is an extremely important parameter that seems to be missing from your understanding of the analysis, which is the WEEK parameter that references the eBird grid data from 2018 to determine the confidence score for the audio clip. As the weeks progress, the confidence score for a particular sound clip will change. That is, an audio clip of a Ruby Throated Humming bird might receive a 98% confidence score during week 30, and a 40% confidence score in week 2 using the same audio clip -- this is because the likelihood of a Ruby-Throated Humming bird is significantly less during week 2 compared to week 40 (that is just an example to illustrate how the WEEK parameter affects detections.

I can assure you that there is nothing in the system that degrades overtime and is somehow replenished from a reboot -- that is at best a coincidence.

I hope that extra piece of very important information increases your understanding of the system and helps you in determining what is happening during your observations.

My best,
Patrick

[jflyons]

Yes that was missing from my understanding, not logically but needed it spelled out I guess! The message regarding the YR Warblers needs to be updated as the detections (over .7) have really mounted up today, as the birds linger. Back to the RUHU example, did the detection algorithm with the week variable changed make the rating go from 45% to 75% with my local yellow-rumps?

This is all very interesting and thanks for explaining patiently. I love this tool and am committed to getting the most out of it for my local population studies. We tend to see what we want to see, and I think many of us expect our tech solutions to degrade so perhaps I will blame that. And of course the reboot is a bit like throwing that pinch of salt over your shoulder or other superstitions!

[jflyons]

Try uploading a low confidence recording to the new model here and see what result you get - https://birdnet.cornell.edu/api/

Thanks @ehpersonal38 - one thing I haven't figured out is how to snag a recording from a non-detected sample, i.e. under .7 threshold?

[ehpersonal38]

Raw recordings are saved to ~/BirdSongs/Processed. It only keeps the most recent 100 though so you'll have to be quick in snagging them!

[jflyons]

A related question? How (or is it possible) does one capture a recording on-demand, if it can? Much like the BirdNET app I am thinking, be able to copy a section of just-heard input, and then analyze? Thanks - just thought I'd put it out there!