Jonas Haag

Unfortunately I don't have a reproducing examle that I can share

``` /home/runner/.conda/envs/X/lib/python3.10/site-packages/sqlalchemy/engine/base.py:1365: in execute return self._exec_driver_sql( /home/runner/.conda/envs/X/lib/python3.10/site-packages/sqlalchemy/engine/base.py:1669: in _exec_driver_sql ret = self._execute_context( /home/runner/.conda/envs/X/lib/python3.10/site-packages/sqlalchemy/engine/base.py:1943: in _execute_context self._handle_dbapi_exception( /home/runner/.conda/envs/X/lib/python3.10/site-packages/sqlalchemy/engine/base.py:2128: in _handle_dbapi_exception util.raise_(exc_info[1], with_traceback=exc_info[2]) /home/runner/.conda/envs/X/lib/python3.10/site-packages/sqlalchemy/util/compat.py:208: in raise_ raise exception /home/runner/.conda/envs/X/lib/python3.10/site-packages/sqlalchemy/engine/base.py:1900: in _execute_context self.dialect.do_execute(...

I recently had this problem too and simply switched to SciPy based STFT, ie. not having the filterbank in PyTorch.

> Then, problem arises when you want to have the iSTFT in the network, for time domain losses, right? Yes but if you put the loss outside the model, no...

I didn't, it is required to be installed at the ONNX export user.

I agree. Do you want to send a PR?

Maybe we can even stop overriding the cache dir. I’m not sure why we do it.

If you’re using Git then I think you are getting “constant” mtimes because Git keeps track of them.

Interesting. Would you mind creating a minimal reproducing example that we could use as a test for that fix?

That would be great! Would be perfect if you could boil it down to using a single file in this repo (or if that doesn’t work, a demo repo)