According to funcwj's uPIT, the training code supporting multi-gpu is written, and the Dataloader is reconstructed.
Demo Pages: Results of pure speech separation model
Generate dataset using create-speaker-mixtures.zip with WSJ0 or TIMI
Prepare scp file(The content of the scp file is "filename path")
python create_scp.py
Prepare cmvn(Cepstral mean and variance normalization (CMVN) is a computationally efficient normalization technique for robust speech recognition.).
#Calculated by the compute_cmvn.py script:
python compute_cmvn.py ./tt_mix.scp ./cmvn.dict
Modify the contents of yaml, mainly to modify the scp address, cmvn address. At the same time, the number of num_spk in run_pit.py is modified.
Training:
sh train.sh
Inference:
sh test.sh