BirdNET-Analyzer and long recordings (1 Viewer)

[OldFatherTime]

I've installed BirdNET-Analyzer on my desktop. How do I use it to analyse long recordings? I understand (I think) that BirdNET handles short clips (around 3 secs) to identify a single bird, so do I have to chop up the long recording myself? This would seem impractical. I've a 2 minute dawn chorus recording in which BirdNET-Analyzer detects just one bird (a Green Woodpecker, which is actually a Song Thrush mimicking one) and does this very quickly, which does make me think it's identifying just a single bird and then stopping.

Out of interest, Merlin makes a good job of identifying all the birds if I play the recording to the app listening on my phone.

 

[LittleBitOfBreadNoCheese]

I've installed BirdNET-Analyzer on my desktop. How do I use it to analyse long recordings? I understand (I think) that BirdNET handles short clips (around 3 secs) to identify a single bird, so do I have to chop up the long recording myself? This would seem impractical. I've a 2 minute dawn chorus recording in which BirdNET-Analyzer detects just one bird (a Green Woodpecker, which is actually a Song Thrush mimicking one) and does this very quickly, which does make me think it's identifying just a single bird and then stopping.

Out of interest, Merlin makes a good job of identifying all the birds if I play the recording to the app listening on my phone.

Point it at the directory containing your files, my recorder produces 2Gb WAV files and birdNET-analyzer will read those fine. You can either use the GUI or a batch file with a command like

BirdNET-Analyzer.exe --i example\MyRecorder --o example\MyRecorder --threads 3 --lat 99.9 --lon -0.22 --week 9

If you have multiple files to do at once you can use multiple threads (I used 3 in the case above).
Make sure you move analyzed files somewhere else when finished so only new files are in the input folder.

 

[OldFatherTime]

Were it that easy. I'm using Linux so that obviously won't work. I was using the gui, but your reply prompted me to look harder for a similar analyze command, and, well, I found that there is a Linux equivalent and it is documented in the installed README.adoc file. RTFM as they say.

It's executed using the python command:

python analyze.py ... (using the same parameters as in the Windows command)

the exact "python" command being however you have python 3.10 installed (being the minimum required at this time).

Thanks for encouraging me to look further!

BTW I find the initial results interesting. BirdNET seems to have problems with the Song Thrush, frequently mis-identifying it as Lesser Spotted Woodpecker.

 

[Vollmeise]

python is my choice also when using BN on my macs, and it's all about workflow when analyzing and annotating tons of recordings.

As You might have noticed, BN results often ar far away from being reliable, even if their confidence is > 0.9.

You can optimize the results by filtering the species list, but in the end You always have to evaluate those results using tools like kaleidoscope or raven pro.

To use kaleidoscope, choose csv in the BN's output format and then adjust the csv-table's column headlines and order manually or e.g. w/ a selfmade python script.

When csv is adjusted, import it in Kaleidoscope and sort the reslts by species then confidence (highest first).

And now it's getting super comfy: tap on suggested species, maybe results of ortolan buntings, and you can scroll in realtime through the dedicated sonograms and evaluate / ID by true / false or own IDs.

When checking 100s of thousands of results this is WAY faster than most other workflows, as You don't have to open any audio file. Just choose a result and kaleidoscope shows the sonogramm of the right place and file.

Another level is using raven pro to evaluate, but adjusting the "raven pro table" exported by BN goes a bit beyond simple magic tricks.

On the other hand. if Your converter script works, Raven Pro is a great tool for e.g. comparing results, customizing sonograms and reading out any audio measurements you need. It's timeline can even show the real time of recording, not only seconds from start of the file.

Feel free to ask if You need any help or imformation.

Cheers)

 

[simpo]

Have you tried Chirpity? It is available for PC, Mac and Linux. You can drop your files or folders onto the app, hit Analyse all and it does the rest. Results appear as they are found and you can click on them to pull up the spectrogram of the call/song at that point in the file. On my Mac it processes at about 130x realtime speed, so more than 2 hours a minute. Results can be filtered by species, sorted by confidence or time, edited as needed and saved to its own library or exprted to CSV, Raven or eBird formats. Its a great piece of software!