Brian McFee

> Actually, python-soxr is using static link for easy installation, so does not need separate libsoxr . I followed that discussion over on the conda-forge PR, but I didn't understand...

Awesome! :tada:

Seconding this: having a handle to the output of one augmenter is crucial if a user wants to apply a cascade of augmentations.

@lostanlen I think to fully support what you describe would require a bit more infrastructure than what I have in mind here, but I agree that it would be great...

Here's a prototype of adaptive specshow: Hack implementation ```python class AdaptiveSpecshow: def __init__(self, interp, mesh): coords = mesh.get_coordinates() self.nx = coords.shape[0] self.ny = coords.shape[1] self.interp = interp self.mesh = mesh...

The implementation above uses linear spacing between min and max for the grid coordinates - probably it should be logarithmic when using log-scaled axes.

```python class AdaptiveSpecshow: def __init__(self, interp, mesh): coords = mesh.get_coordinates() self.nx = coords.shape[0] self.ny = coords.shape[1] self.interp = interp self.mesh = mesh self.cidx = None self.cidy = None self.ax =...

It turns out that we can do the coordinate interpolation in axes space (0, 1) and then use the axis transform inverses to map back into data coordinates. This naturally...

Ok, I've tracked down the issue with tempogram and inverted plots. Resolving this in the "right" way will bump our minimum scipy dependency up to 1.9, so I'm clearing it...

This could make it back onto the 0.10 roadmap if/when scipy 1.9 is released.