Rachel Bittner

cc @lostanlen @magdalenafuentes Hey @aahlenst ! We're also open to collaboration. For some context, the goal of [mirdata](https://github.com/mir-dataset-loaders/mirdata) is to act a bit like sklearn.datasets but for music. mirdata is...

Hey @lflee the `predict_on_audio.py` script is currently written to output in one of two formats -- salience (the npz files), multif0 csv files with time stamps and multiple frequency values...

``` import numpy as np # load an npz file created by predict_on_audio.py data_dictionary = np.load("mynpy_file.npz") times = data_dictionary['times'] freqs = data_dictionary['freqs'] salience = data_dictionary['salience'] ``` `salience` is the `gram`...

@urinieto what other defaults did you see besides the 0.5 that you think need to be abstracted out?

The parameters you mention aren't so important for melody extraction. (For ex. the overlap threshold has almost no effect). If I understand right though, the larger issue you're trying to...

> Agreed with 1. We need green light from @rabitt for 2. I'm OK with a 1.5.0 release to sync the two. I have on my backlog to set up...

Hey @shakfu > My question to @rabitt is whether it is possible to translate this script's functionality (in particular the ": newfile : restart" idiom) into a pure-python pysox solution....

@nefastosaturo thanks for catching this! If you'd like, feel free to open a PR with the fix you propose.

@semins Yep, you're right, it should accept negative values. Feel free to make a PR if you'd like!

Hey @cristianritter - its admittedly a bit confusing, but there are logs coming from two different places. `tfm.set_globals(verbosity=0)` controls the sox command line's verbosity, but the warning you're seeing is...