Hi all. Hope everyone is doing fine.
The problem
I always knew that, for my project, and just for a small proof of concept, data was going to be a problem.
I already did record a number of birds - mainly house sparrows - with my mobile phone, but the quality was not the best. It's not that the audio is not fine, it is, but the bird calls are not at the expected definition for a model to learn.
I was going to try my best to record more and more sounds and see if the model could learn and identify them .
A Solution
While searching for some tutorials on audio, I stumble across BirdNET.
What is BirdNet.
BirdNET is a project from the K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology and the Chair of Media Informatics and Chemnitz University of Technology that tries to answer the question - How can computers learn to recognize birds from sounds ?
The main focus of the research is the detection and classification of avian sounds using Machine Learning. BirdNet it's a citizen science platform. They have apps for IOS and Android, as well supporting various hardware and software platforms, such as Arduino, Raspberry PI, web browsers, IOS and Android - but no RP2040. 
They have all the code in Github, including the ML models.
Not reinventing the wheel
Because my main problem was data, the project, by now, supports around 3,000 of the world's most common birds - mainly from USA and Europe.
Now, I will not have that problem.
I'll solely focus on getting BirdNET (they already have a tflite INT8 model ) working on a Raspberry PICO.
Stay tuned for further developments.