This PR fixes #62, adding the long-requested replay function.
Given the JAMS output from a previous run of muda, and a freshly jam-packed input, replay will apply the exact transformation history of the previous run to the new input. This is useful for reconstructing the deformed audio from a deformed annotation and the clean audio.
This, in turn, makes it possible to share the outputs of muda pipelines efficiently, so that any practitioner with a clean copy of the data can reconstruct deformation outputs independently, and with neither dependence on the exact muda object and randomstate, or the entire audio output, which can be too large to share in practice.
This PR fixes #62, adding the long-requested
replayfunction.Given the JAMS output from a previous run of muda, and a freshly jam-packed input, replay will apply the exact transformation history of the previous run to the new input. This is useful for reconstructing the deformed audio from a deformed annotation and the clean audio.
This, in turn, makes it possible to share the outputs of muda pipelines efficiently, so that any practitioner with a clean copy of the data can reconstruct deformation outputs independently, and with neither dependence on the exact muda object and randomstate, or the entire audio output, which can be too large to share in practice.
This change is