In the process of installing BirdVoxDetect in a Python 3.7 environment, pip downloaded (or retrieved from local cache) no fewer than eight versions of TensorFlow, beginning with 2.6.0 and decreasing in version number to 2.3.4, which it finally settled on. A similar thing happened with SciPy. I noticed this especially since those packages are relatively large and I happened to be using a slow internet connection. Perhaps an adjustment to some of the dependency version number requirements in your setup.py would help with this issue? I notice, for example, that you require NumPy 1.16.4, which is more than two years old now.
Relevant context: when analyzing BV-300h and BVFS i have used these versions
From Harold Mills:
Relevant context: when analyzing BV-300h and BVFS i have used these versions
Solution: set an upper bound to tensorflow dependency, namely 2.3.4 (thanks Harold!)
Triaged for v0.6
For v0.7 we'll explore the possibility of using more recent versions of numpy and scipy. but that's a separate question