echocatzh / mtfaa-net Goto Github PK

thanks for your code, there is a problem still confuse me, the input of the u-net structure is the magnitude after the phase encoder, but the output of the u-net have two-stage mask, one is magnitude mask, the other is phase mask and magnitude mask, I am confusing that there is no phase information input to the u-net structure, how can it get the correct phase mask? or after phase encoder, although the output is magnitude, but it includes phase information?

论文中MTFAA-Net-Streaming的lookahead有40ms，我没有在代码中找到具体的体现，论文的帧移是8ms，应该是用了5帧的未来帧，在哪里体现呢？

Hi,

How can I train a model with my dataset. Where can I find sample usage for training the model?

Thank you

Hi! Iwonder if we can reduce nerb which is currently using 256. I think if it has less number of bands it can reduce IOPS.

based on Baidu paper, I guess they are using ERB Bands applied in [10].

As I noticed in PercepNet paper(A Perceptually-Motivated Approach for Low-Complexity, Real-Time
Enhancement of Fullband Speech) they are using 34 bands with 20khz highest frequency.

Hi, thanks for you great work.
I find that the loss decrease hardly when I train your MTFAA, I dont normalize signals when I calculate loss.
Maybe I should normalize signals like 《Data augmentation and loss normalization for deep noise suppression》，I want to know your way to calculate loss.

Did you test the model's real performance on real AEC data? and what's the FLOPS and parameters？

Hello,

First of all, thank you for proving the implementation. It was very helpful to understand the paper.

I had one question though. When I was trying to train the network using 30-second 48kHz audio, I always run into CUDA out of memory error, even if the batch size is set to 1. Have you seen that in your experiments or do you have any advice maybe?

Anything will be greatly appreciated!

The report is as follows:

  File "E:/code_paper/MTFAA-Net-main/erb.py", line 24, in __init__
    filter = th.from_numpy(filter).float()
TypeError: expected np.ndarray (got tuple)

The error occurred on line 24 of erb.py.
filter = th.from_numpy(filter).float()
"filter" is a tuple has two members.

My Python version is 3.9.12.
My spafe version is 0.2.0.
My torch version is 1.12.1.

Hope you can teach me.

你好，大佬，感谢开源这么优秀的项目；看到论文中说“LAEC中引入附加的条件信息，可以进一步提高模型对回声任务的性能。但是，如果将LAEC与模型简单地连接在一起，由于LAEC引入的失真会降低系统的性能”，有几个问题想请教下呢：
1、这个模型输入是三个吧，混合音频、LACE数据、远端数据
2、这个LAEC的输出要经过的附加条件是指什么呢，不太理解这个，LAEC与模型直接连接，是不是就是将LAEC的输出直接和混合音频、远端数据一起送到模型呢。求大佬指教

Hi @echocatzh

I think there are many people supposed to use your awesome work for both commercial and non-commercial purposes
it would be great for people who use this and of course for you as well if you could add an explicit License
would you be able to add the license file?