tky823 / ssspy Goto Github PK

• Lee, T. Kim, and T.-W. Lee, "Fast fixed-point independent vector analysis algorithms for convolutive blind source separation," Signal Processing, vol. 87, no. 8, pp. 1859–1871, 2007.

• N. Ono and S. Miyabe, "Auxiliary-function-based independent component analysis for super-Gaussian sources," in Proc. LVA/ICA, pp.165–172, 2010.

Documentation is important to users.
Now, I haven't installed document tests, but I want to do it.

This issue includes measure of document coverage.

MNMF and its derivations

• MNMF (H. Sawada, H. Kameoka, S. Araki, and N. Ueda, "Multichannel extensions of non-negative matrix factorization with complex-valued data," IEEE Trans. ASLP, vol. 21, no. 5, pp. 971-982, 2013)
• t-MNMF
• FastMNMF
  • See #35

A sample dataset is necessary to verify separation methods correctly work.
This function is also required to tests/.

Pull request template helps us open PR in a standardized format.

AuxIn IVA-ISS, demixing filters are not computed, so _reset will result in failure.

np.random.rand is legacy style.
Use random number generator like:

rng = np.random.default_rng()
rng.random(size=(5, 2))

• D. Kitamura et al., "Determined blind source separation unifying independent vector analysis and nonnegative matrix factorization," IEEE/ACM Trans. ASLP., vol. 24, no. 9, pp. 1626-1641, 2016.
• S. Mogami et al., "Independent low-rank matrix analysis based on complex student's t-distribution for blind audio source separation," in Proc of MLSP, 2017.
• D. Kitamura et al., "Generalized independent low-rank matrix analysis using heavy-tailed distributions for blind source separation," pp. 1-25, 2018.

• A. Brendel and W. Kellermann, "Faster IVA: Update rules for independent vector analysis based on negentropy and the majorize-minimize principle,"

• H. Sawada, S. Araki, and S. Makino, "Underdetermined convolutive blind source separation via frequency bin-wise clustering and permutation alignment," in IEEE Trans. ASLP, vol. 19, no. 3, pp. 516-527, 2010.

It is useful to add extras_require option to setup.cfg for dev and docs.

• K. Yatabe and D. Kitamura, "Determined blind source separation via proximal splitting algorithm," in Proc. ICASSP, 2018, pp. 776-780.

Auxiliary-function-based demixing filter update is common in FDICA, IVA, and ILRMA.
We can create submodule or function to handle this.

Create ssspy.linalg.eigh by using NumPy.

• papers or agorithms
  • R. Ikeshita, "Independent positive semidefinite tensor analysis in blind source separation," in Proc. EUSIPCO, 2018, pp. 1652-1656.
  • T. Kondo, K. Fukushige, N. Takamune, D. Kitamura, H. Saruwatari, R. Ikeshita, and T. Nakatani, "Convergence-guaranteed independent positive semidefinite tensor analysis based on Student's t distribution," in Proc. ICASSP, 2020, pp. 681-685.
  • MM-algorithm based Gauss-IPSDTA
• linear algebra
  • principal square root of semidefinite matrix
  • inversion of principal square root of semidefinite matrix

The links to each notebook will be helpful.
Write links in README.md like

[![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/tky823/ssspy/blob/main/notebooks/bss/GaussILRMA-IP1.ipynb)

• N. Ono, "Auxiliary-function-based independent vector analysis with power of vector-norm type weighting functions," in Proc of APSIPA ASC, 2012.

Now, code coverages are updated multiple times. This is time-consuming.

The separation quality of GaussILRMA-IP2 is limited.
I have to investigate the cause.

• R. Scheibler and N. Ono, "Fast and stable blind source separation withrank-1 updates," in Proc. ICASSP, pp. 236–240, 2020.
• R. Ikeshita and T. Nakatani, "ISS2: An Extension of Iterative source steering algorithm for majorization-minimization-based independent vector analysis," arXiv preprint arXiv:2202.00875, 2012

List up implemented methods.

• T. Kim et al., "Blind source separation exploiting higher-order frequency dependencies," IEEE Trans. ASLP., vol. 15, no. 1, pp. 70–79, 2007.

The arguments should_<do>_* are too long.
How about use_* instead?

Now, we define callback functions and classes in each test script.

• A. Hyvärinen, "Fast and Robust Fixed-Point Algorithms for Independent Component Analysis," IEEE Trans. on Neural Netw., vol. 10, no. 3, pp. 626-634, 1999.

• N. Murata, S. Ikeda, and A. Ziehe, "Minimal distortion principle for blind source separation," in Proc. ICA, 2001, pp. 722-727.

I have to add a test for 4 sources separation.

Auxiliary-function-based FDICA should be tested.

The test is important to keep the quality of code high.
However, it takes a lot of time to test codes.

Practical solutions

• Shorter input signal
• Fewer iterations

Summary

• Fast Gauss-MNMF
• Fast t-MNMF
• Fast GGD-MNMF

Now, ssspy selects updating pair using itertools.combinations(range(self.n_sources), 2).
This behavior is different from papers.

ssspy.transform.whiten only supports complex inputs.

np.linalg.slogdet(X) is more robust than np.log(np.linalg.det(X))

• Dynamic independent component/vector analysis: time-variant linear mixtures separable by time-invariant beamformers

ssspy.bss._update_spatial_model:update_by_ip2 returns different result from old one.

Now, the permutation solver is based on the correlation between frequencies.
Degree-of-arrival based method may be useful.

• S. Ikeda, and M. Noboru, "A method of ICA in time-frequency domain." in Proc. ICA, 1999.

The settings of code-cov are in default.
Some of the settings may be changed or customized.

In FastICA, FastIVA, and FasterIVA, the input should be whitened before the fixed-point-iteration-based algorithm is applied.
However, the timing of whitening is different among these methods.

• N.Ono, "Stable and fast update rules for independent vector analysis based on auxiliary function technique," in Proc. WASPAA, pp.189–192, 2011.

• Principal component analysis
• Whitening (a.k.a sphering)

Reference
https://github.com/tky823/ssspy/projects/1#card-81816570

isort reorders the python modules, which improves the intelligibility of this library.

Summary

Jupyter notebooks may help users understand how to utilize this library.

Notebooks

ICA

• Gradient-descent-based ICA
• Natural-gradient-descent-based ICA
• FastICA

FDICA

• Gradient-descent-based FDICA
• Natural-gradient-descent-based FDICA
• Auxiliary-function-based FDICA (IP1)
• Auxiliary-function-based FDICA (IP2)

IVA

• Gradient-descent-based IVA
• Natural-gradient-descent-based IVA
• FastIVA
• FasterIVA
• Auxiliary-function-based IVA (IP1)
• Auxiliary-function-based IVA (IP2)
• Auxiliary-function-based IVA (ISS1)
• Auxiliary-function-based IVA (ISS2)

ILRMA

• Gauss-ILRMA (IP1)
• Gauss-ILRMA (IP2)
• Gauss-ILRMA (ISS1)
• Gauss-ILRMA (ISS2)
• t-ILRMA (IP1)
• t-ILRMA (IP2)
• t-ILRMA (ISS1)
• t-ILRMA (ISS2)
• GGD-ILRMA (IP1)
• GGD-ILRMA (IP2)
• GGD-ILRMA (ISS1)
• GGD-ILRMA (ISS2)