tky823
tky823
I have not read the constraints, so I will have a look. I've confirmed that DANet separates the mixtures. But I have not checked the score (e.g. SDR improvement). As...
| Model | Size (in original paper)| Size (mine) | |:-:|:-:|:-:| | Deep clustering | | | | DPTNet | 2.69M | 2.8M |
I added these metrics (SDR, SIR, and SAR) using `mir_eval` package. This evaluation considers the time-invariant filter, but the mixture is assumed as an instantaneous mixture. Should I implement it...
Hi, thank you for sharing your codes. I got 2.632M (2,631,680) model size using your code. Do the hyperparameters match the configuration in the [paper](https://arxiv.org/abs/2007.13975)? In the paper, the model...
Using the following script, I got different values: 2.835M (2,834,945) `thop` library may miss some trainable parameters. ```python # dpt_net.py def main(): model = DPTNet_base(enc_dim=256, feature_dim=64, hidden_dim=128, layer=6, segment_size=250, nspk=2,...
Fixed in MNMF.
In my naive implementation, FastIP is not faster than conventional IP. Rather, FastIP is a little bit slower. https://colab.research.google.com/github/tky823/ssspy/blob/cc398f53a16a99591ee65a8d236fd4f062209534/notebooks/ISSUE/AuxIVA-FastIP1.ipynb https://colab.research.google.com/github/tky823/ssspy/blob/cc398f53a16a99591ee65a8d236fd4f062209534/notebooks/ISSUE/GaussILRMA-FastIP1.ipynb
I think the priority is low, so move milestone to `v0.3.0`.
```math \begin{align} \boldsymbol{\Pi}_{n} &= (\bar{\boldsymbol{E}}_{n}; \boldsymbol{e}_{n}) \\ \boldsymbol{\Pi}_{n}\tilde{\boldsymbol{U}}_{in}\boldsymbol{\Pi}_{n}^{\mathsf{T}} &= \left( \begin{array}{cc} \boldsymbol{\Phi}_{in} & \boldsymbol{\mu}_{in} \\ \boldsymbol{\mu}_{in}^{\mathsf{H}} & \eta_{in} \end{array} \right) \end{align} ``` ```math \left\{(\boldsymbol{\Pi}_{n}\tilde{\boldsymbol{U}}_{in}\boldsymbol{\Pi}_{n}^{\mathsf{T}})^{*}\right\}^{-1} ={\displaystyle \left( \begin{array}{cc} \left(\boldsymbol{\Phi}_{in}^{*} - \dfrac{\boldsymbol{\mu}_{in}^{*}\boldsymbol{\mu}_{in}^{\mathsf{T}}}{s_{in}}\right)^{-1}...