-
As psd_n and psd_s are Hermitian and symmetric matrices, can torch.linalg.eig() be replaced with torch.linalg.eigh() in _get_steering_vector_evd() for faster computes the eigenvalue decomposition?
-
As psd_n and psd_s are Hermitian and symmetric matrices, can torch.linalg.eig() be replaced with torch.linalg.eigh() in _get_steering_vector_evd() for faster computes the eigenvalue decomposition?
-
### 🐛 Describe the bug
the return of torchaudio.transforms.MVDR.forward() should be _return specgram_enhanced.to(dtype)_ rather than _return specgram_enhanced_, which mght cause dtype inconsistency b…
-
`mvdr` is regisered and described as `MVDDepthResolutionBox`. However, in Part 15 there is no `mvdr` defined and `MVDDepthResolutionBox` is defined as:
```cpp
class MVDDepthResolutionBox extends B…
-
Hi there, thanks for your great work.
I just updated my local code following the PR #3838 (add the loss criterion and the wrapper), and find that the performance degrades on both sdr and asr perfo…
-
./run.sh --id 0,1,2,3 --stage 3
Results from the following experiment will be stored in exp/train_convtasnet_reverb2reverb_8kmin_e2ff7662
Stage 3: Training
C:\Users\SASPL-1\anaconda3\envs\zx_38\lib…
-
https://github.com/funcwj/aps/blob/3f46e4750923a3c50529403614eda11118b53b9a/aps/asr/filter/mvdr.py#L57
您好,我想问一下,这行代码是MVDR的分子部分,应该是mask之后的特征和其共轭转置矩阵乘,您的写法后面使用的是mask之前的特征,不会有问题嘛?
-
### 🐛 Describe the bug
NCCL currently only supports floating point and integer dtype. (See https://github.com/pytorch/pytorch/issues/45760)
The MVDR module forces to use ``cdouble`` dtype to improve…
-
### Description
cx using the example got error
https://learn.microsoft.com/en-us/powershell/module/az.reservations/get-azreservationquote?view=azps-9.1.0#examples
Get-AzReservationQuote -AppliedS…
-
Is it possible to run https://colab.research.google.com/drive/1UVoYDUiIrwMpBTghQPbA6rC1mc9IBzi6?usp=sharing with GPU?
I tried to move data to cuda via `.to(cuda:0)` but received
```
File "/senv/li…