tky823

Implementation of D3Net

11

Reference: ["D3Net: Densely connected multidilated DenseNet for music source separation"](https://arxiv.org/abs/2010.01733)

The number of parameter is different.

1

In deep clustering, the number of trainable parameters is different from [here](https://arxiv.org/abs/1809.07454).

In the [DANet paper](https://arxiv.org/abs/1611.08930), the curriculum training was used.

Evaluation metrics

1

Evaluation of source separation by - SDR (improvement) - SIR (improvement) - SAR These are realized by `mir_eval`.

### 🐛 Describe the bug In an example of https://pytorch.org/audio/stable/transforms.html, `TimeStretch` takes arguments as follows: ```python TimeStretch(stretch_factor, fixed_rate=True) ``` This usage is incorrect. According to https://pytorch.org/audio/stable/generated/torchaudio.transforms.TimeStretch.html, this class takes the...

- [ ] SparseAuxIVA ["A computationally cheaper method for blind speech separation based on AuxIVA and incomplete demixing transform"](https://ieeexplore.ieee.org/document/7602921) - [ ] SparseProxIVA ["Time-frequency-masking-based Determined BSS with Application to Sparse...

Independent Positive Semidefinite Tensor Analysis - ["Independent Positive Semidefinite Tensor Analysis in Blind Source Separation"](https://ieeexplore.ieee.org/document/8553546)

- [x] t-ILRMA - [ ] GGD-ILRMA - [ ] KL-ILRMA - [x] ~Consistent-ILRMA~

Initialization of partitioning function

1

Value of `z` is initialized randomly at ILRMA, MNMF

- [x] ~Complex NMF~ - [x] ~t-NMF~ - [x] ~Cauchy-NMF~ - [ ] NMF deconvolution ME algorithm based NMF - [x] ~IS-NMF~ - [x] ~Cauchy-NMF~