Home

Awesome

dsdtools

:warning: We released MUSDB18 our newer, larger dataset. Please use the musdb python package to parse it.

Build Status Coverage Status Docs Status

A python package to parse and process the demixing secrets dataset (DSD) as part of the MUS task of the Signal Separation Evaluation Campaign (SISEC).

Installation

pip install dsdtools

DSD100 Dataset / Subset

The complete dataset (~14 GB) can be downloaded here. For testing and development we provide a subset of the DSD100 for direct download here. It has the same file and folder structure as well as the same audio file formats but consists of only 4 tracks of 30s each.

Usage

This package should nicely integrate with your existing python code, thus makes it easy to participate in the SISEC MUS tasks. The core of this package is calling a user-provided function that separates the mixtures from the DSD into several estimated target sources.

Providing a compatible function

Here is an example for such a function separating the mixture into a vocals and accompaniment track:

def my_function(track):
    # get the audio mixture as
    # numpy array shape=(nun_sampl, 2)
    track.audio

    # compute voc_array, acc_array
    # ...

    return {
        'vocals': voc_array,
        'accompaniment': acc_array
    }

Creating estimates for SiSEC evaluation

Setting up dsdtools

Simply import the dsdtools package in your main python function:

import dsdtools

dsd = dsdtools.DB(root_dir='path/to/dsdtools')

The root_dir is the path to the dsdtools dataset folder. Instead of root_dir it can also be set system-wide. Just export DSD_PATH=/path/to/dsdtools inside your terminal environment.

Test if your separation function generates valid output

Before processing the full DSD100 which might take very long, participants can test their separation function by running:

dsd.test(my_function)

This test makes sure the user provided output is compatible to the dsdtools framework. The function returns True if the test succeeds.

Processing the full DSD100

To process all 100 DSD tracks and saves the results to the folder estimates_dir:

dsd.run(my_function, estimates_dir="path/to/estimates")

Processing training and testing subsets separately

Algorithms which make use of machine learning techniques can use the training subset and then apply the algorithm on the test data. That way it is possible to apply different user functions for both datasets.

dsd.run(my_training_function, subsets="Dev")
dsd.run(my_test_function, subsets="Test")
Access the reference signals / targets

For supervised learning you can use the provided reference sources by loading the track.targets dictionary. E.g. to access the vocal reference from a track:

track.targets['vocals'].audio

If you want to exclude tracks from the training you can specify track ids as the dsdtools.DB(..., valid_ids=[1, 2]) object. Those tracks are then not included in Dev but are returned for subsets="Valid".

Processing single or multiple DSD100 tracks

dsd.run(my_function, ids=30)
dsd.run(my_function, ids=[1, 2, 3])
dsd.run(my_function, ids=range(90, 99))

Note, that the provided list of ids can be overridden if the user sets a terminal environment variable DSD_ID=1.

Use multiple cores

Python Multiprocessing

To speed up the processing, run can make use of multiple CPUs:

dsd.run(my_function, parallel=True, cpus=4)

Note: We use the python builtin multiprocessing package, which sometimes is unable to parallelize the user provided function to PicklingError.

GNU Parallel

GNU parallel is a shell tool for executing jobs in parallel using one or more computers. A job can be a single command or a small script that has to be run for each of the lines in the input. The typical input is a list of files, a list of hosts, a list of users, a list of URLs, or a list of tables. A job can also be a command that reads from a pipe. GNU parallel can then split the input and pipe it into commands in parallel.

By running only one id in each python process the DSD100 set can easily be processed with GNU parallel using multiple CPUs without any further modifications to your code:

parallel --bar 'DSD_ID={0} python main.py' ::: {0..99}

Compute the bss_eval measures

The official SISEC evaluation relies on MATLAB because currently there does not exist a bss_eval implementation for python which produces identical results. Therefore please run dsd100_eval_only.m from the DSD100 Matlab scripts after you have processed and saved your estimates with dsdtools.

Full code Example

import dsdtools

def my_function(track):
    '''My fancy BSS algorithm'''

    # get the audio mixture as numpy array shape=(num_sampl, 2)
    track.audio

    # get the mixture path for external processing
    track.path

    # get the sample rate
    track.rate

    # return any number of targets
    estimates = {
        'vocals': vocals_array,
        'accompaniment': acc_array,
    }
    return estimates

# initiate dsdtools
dsd = dsdtools.DB(root_dir="./Volumes/Data/dsdtools")

# verify if my_function works correctly
if dsd.test(my_function):
    print "my_function is valid"

# this might take 3 days to finish
dsd.run(my_function, estimates_dir="path/to/estimates")

References

If you use this package, please reference the following paper

@inproceedings{
  SiSEC17,
  Title = {The 2016 Signal Separation Evaluation Campaign},
  Address = {Cham},
  Author = {Liutkus, Antoine and St{\"o}ter, Fabian-Robert and Rafii, Zafar and Kitamura, Daichi and Rivet, Bertrand and Ito, Nobutaka and Ono, Nobutaka and Fontecave, Julie},
  Editor = {Tichavsk{\'y}, Petr and Babaie-Zadeh, Massoud and Michel, Olivier J.J. and Thirion-Moreau, Nad{\`e}ge},
  Pages = {323--332},
  Publisher = {Springer International Publishing},
  Year = {2017},
  booktitle = {Latent Variable Analysis and Signal Separation - 12th International Conference, {LVA/ICA} 2015, Liberec, Czech Republic, August 25-28, 2015, Proceedings},
}