Speech separation project

Aksenov Yaroslav

Installation guide

Download model to the folder final_model

pip install -r ./requirements.txt

Launching guide

Generating data

Create folder data/datasets/speech_separation/, where the dataset will be created using command:

python src/mixer/generate_ss_data.py

Testing:

python test.py \
   -c final_model/test_config.json \
   -r final_model/model_best.pth \
   -t TEST_DATASET_PATH \
   -o test_result.json

Training step 1:

python train.py \
   -c src/train_config_step1.json \
   -wk "YOUR_WANDB_API_KEY"

Training step 2:

python train.py \
   -c src/train_config_step2.json \
   -wk "YOUR_WANDB_API_KEY" \
   -p "PATH TO THE LATEST CHECKPOINT FROM TRAIN 1"

Training step 3:

python train.py \
   -c src/train_config_step3.json \
   -wk "YOUR_WANDB_API_KEY" \
   -p "PATH TO THE LATEST CHECKPOINT FROM TRAIN 2"