search

Andong-Li-speech / EaBNet

This is the repo of the manuscript "Embedding and Beamforming: All-Neural Causal Beamformer for Multichannel Speech Enhancement", which was submitted to ICASSP2022.
78 stars 15 forks source link

Data Mixing #2

Closed hido710 closed 2 years ago

hido710 commented 2 years ago

I am amazed by your results!

Actually, after finding out that the pre-trained models were not in the directory, I tried to train the models myself. But, the data mixing stages are a bit confusing for me.

  1. Did you locate the microphone array with parallel to room width or length? or with definitely random angle?
  2. Did you locate the sources with 0 to 180 degrees from the center of microphone array? or 0 to 360 degrees (so that the target source and the interference can be placed on the opposite side of the microphone array.)?

Could you please answer this question or share the reference code for mixing the data?

Hope you have a great day, bye!

Andong-Li-speech commented 2 years ago

I am amazed by your results!

Actually, after finding out that the pre-trained models were not in the directory, I tried to train the models myself. But, the data mixing stages are a bit confusing for me.

  1. Did you locate the microphone array with parallel to room width or length? or with definitely random angle?
  2. Did you locate the sources with 0 to 180 degrees from the center of microphone array? or 0 to 360 degrees (so that the target source and the interference can be placed on the opposite side of the microphone array.)?

Could you please answer this question or share the reference code for mixing the data?

Hope you have a great day, bye!

  1. The ULA is parallel to room length without rotation in the simulation process.
  2. As we adopt the ULA in the simulation, only 0 to 180 degrees are considered. If you use the circular array in the simulation, you can extend the angle range into [0, 360) or [-180, 180).
hido710 commented 2 years ago

Thanks for kindly reply. Have a good day!