Python Tensorflow and it's tf_runtime mostly have problems keeping up with new python versions (like most scientific libs). Would it not be more elegant to rely on "Tensorflow Lite C API" instead? However there are special operators within the birdnet models (e.g. BirdNET_6K_GLOBAL_MODEL.tflite) for FFT or Mel Spectrograms, which would need a library like "ltensorflowlite_flex", which I find quite complicated to compile using bazel and the like. But I think, it would be worth it and could avoid running several python versions in parallel, no?
Python Tensorflow and it's tf_runtime mostly have problems keeping up with new python versions (like most scientific libs). Would it not be more elegant to rely on "Tensorflow Lite C API" instead? However there are special operators within the birdnet models (e.g. BirdNET_6K_GLOBAL_MODEL.tflite) for FFT or Mel Spectrograms, which would need a library like "ltensorflowlite_flex", which I find quite complicated to compile using bazel and the like. But I think, it would be worth it and could avoid running several python versions in parallel, no?