Tags: descriptinc/audiotools
Tags
Prepping for open source. (#80) * Check if is cuda. * Adding stuff. * no more print statements * Deleting outdated tools for release. * Bringing back post. * bringing back tests * Adding preference test tools + tests * Adding to tests. * Adding again. * Adding stuff from Hugo's PR. * Fixing for tests, adding docs. * Adding timeline notches + dividing line. * Adjusting color. * Make cookie name customizable. * Version bump. --------- Co-authored-by: pseeth <prem@descript.com>
Adding decorators for training progress bars, etc. (#79) * Adding decorators * Rewrote decorators. * Adding demo * Added test. * Added the decorators. * Added docstring * Bumping version. * Default step_fn * Explicitly set tracker step. * Moving step * Adding step to init * dict, not defaultdict when saving history * Moving lambda outside. * Changing torchmetrics. * making dummy * Device * return self * missing item * No more torchmetrics * Load step * Allow writer to be None * Moving write call * Moving ddp active to arguments of track. * Longer width * skip nans in taking mean. * Fixing mean. * Cleaning up decorators a bit. * Adding docstrings * Should update even if not dict, for pbars. * Fixing issue with dropping float/int. --------- Co-authored-by: pseeth <prem@descript.com>
loader should sample from files without replacement (#73) * loader should sample from files without replacement * lint * codecov * lint * lint * fixing lint * Adding docstrings --------- Co-authored-by: Hugo Flores Garcia <hfgarcia@descript.com> Co-authored-by: pseeth <prem@descript.com>
Padding amount was too high for loading files with FFMPEG. (#72) * Getting the pad amount as low as possible for mp3 to pass. * Fleshing out comment. * Bumping version * Raising threshold. * Adding source * updating comment * Adding offset option to dataset * Changing to hann window * Update regression data. * Raising atol but why --------- Co-authored-by: pseeth <prem@descript.com>
Refactoring the data pipeline. (#71) * Got the main stuff done, still a bunch of tests to write. * propagate find_audio args * Coverage on datasets. * Fixing test, adding docs. * Adding to find_audio test. * Fixing train_separator script * Fixing * Fixing sisdr loss and train_separator. * Examples working * Rewriting how things get aligned. * Bigger version bump * Adding test * Adding max history to autoclip * Adding something for gradient accumulation in the collater. * Fixing comment * Removing comment --------- Co-authored-by: pseeth <prem@descript.com>
PreviousNext