BirdNET instruction

Sunny Tseng 2023-04-17

Pre-requisite

Google Colab (https://colab.research.google.com/)

You will need to sign in to Google Colab to get access to the computational resources provided by Google. If you have a Gmail account, you are all set. Enter the Google Colab webpage, then click the blue button Sign in at the upper right corner. Enter your Gmail and password, and you should be brought to the Colab main page.

Google Drive (https://www.google.com/intl/en_ca/drive/)

Here is where you are going to store the audio files that going to be analyzed by BirdNET. Create a folder called BirdNET_WAV, then upload your audio file(s) into the folder.

Python Notebook

Download the python script BirdNET_Lite_JPRF.ipynb that is provided in this Github repository. Or, you can click on the Open in Colab botton once you open the ipynb file on Github.

Analysis

• Open the python notebook in Google Colab by clicking File -> Open notebook -> Upload, then locate or drag & drop the python notebook file that you just downloaded.

• Run the first two chunks of code. You can do this by clicking the arrow at the top left corner of the code chunk, or by entering Ctrl + Enter - the first chunk is to load the libraries, and the second chunk is to connect your Google drive with the Colab remote server, and also download the BirdNET algorithm to the remote server. Here is what you should see if everything goes well:

Mounted at /content/gdrive
Cloning into 'BirdNET-Lite'...
remote: Enumerating objects: 29, done.
remote: Total 29 (delta 0), reused 0 (delta 0), pack-reused 29
Unpacking objects: 100% (29/29), 58.09 MiB | 10.00 MiB/s, done.

• Click on the folder icon on the left panel - you should see the file structure in your remote server, where BirdNET-Lite and gdrive should both appear. The BirdNET-Lite folder includes all the identification algorithms and the gdrive is basically your Google Drive, which is now a folder in the remote server.

• Run the third chunk of the code to change the current working directory to BirdNET-Lite.

• Now we are at the firth chunk of the code, where the main analysis is happening. The code basically loop through all the .wav file in the BirdNET_WAV folder then produce outcome in a newly created folder called BirdNET_WAV_output in your Google Drive (the code will create this folder for you so you don't need to do it by yourself).

• Run the forth chunk of the code. You should see something like this if you succeed :) This might take a while depending on how many recordings you have. Each 1-mins audio recording would probably takes about 5-secs to analyze.

INFO: Created TensorFlow Lite delegate for select TF ops.
INFO: TfLiteFlexDelegate delegate: 1 nodes delegated out of 182 nodes with 1 partitions.

INFO: Created TensorFlow Lite XNNPACK delegate for CPU.
LOADING TF LITE MODEL... DONE!
READING AUDIO DATA... DONE! READ 37 CHUNKS.
ANALYZING AUDIO... DONE! Time 5.0 SECONDS
WRITING RESULTS TO BirdNET_output.txt ... DONE! WROTE 5 RESULTS.

• If you are still following, congrats! Now to look up your result, go to your Google Drive and find the BirdNET_WAV_output folder. You should see the many .csv files there. Each .csv file corresponds to each sound recording. You should be able to see which birds are included in the recording by reading through the .csv file.

Combine all .csv files into one metadata

• To combine the .csv files into only one file, download the BirdNET_WAV_output into your computer. Put the folder at the same working directory with the R code group_BirdNET_output.R.

• Run the the group_BirdNET_output.R in RStudio then you will get the final combined object, which is called data_full_format. Here is how the data should look like:

Well done :)