Skip to content

Time-domain Audio Separation Network (IN PYTORCH)

Notifications You must be signed in to change notification settings

runninging/TASNET

Repository files navigation

TASNET: Time-domain Audio Separation Network

Two models are provided here, one with SDR = 11.7 and the other with SDR = 13.6

Model with SDR = 11.7

You can find the oringnal paper here. The testing results are shown as follows: SDR-11.7

Function

  • dataset.py: read the data into the model
  • Tasnet_model.py:the forword network
  • Tasnet_train.py:the main function to run
  • trainer.py:calculate the loss and for training and testing
  • utils.py: process the raw audio and other useful functions
  • train.yaml: all the parameters used in the model
  • test.py:separate the mixed audio and calculate SDR

Training stage:

  • from the beginning: remove the line with "trainer.rerun" in Tasnet_train.py, use "trainer.run" instead
  • from a trained model: remove the line with "trainer.run" in Tasnet_train.py,use "trainer.rerun" instead, and change the "model_path" in train.yaml/temp

Model with SDR = 13.6

You can find the oringnal paper here.

Function

  • Tasnet_model_13.6.py: the forword network of SDR = 13.6 system

The only diffierence between 11.7 and 13.6 system is the forward network, just replace the Tasnet_model.py with Tasnet_model_13.6.py and run.

Releases

No releases published

Packages

No packages published

Languages