Course design of signal processing, by jineng han, 2020-06-10
Required python libraries: Tensorflow with GPU support (>=1.4) + Scipy (>=1.1) + Numpy (>=1.14) + Tqdm (>=4.0.0). To install in your python distribution, run
pip install -r requirements.txt
Required software (for resampling): SoX
To convert audiofile.wav to 32-bit floating-point audio at 16kHz sampling rate, run:
sox audiofile.wav -r 16000 -b 32 -e float audiofile-float.wav
If you just want to test the model, please download the default validation data by running:
./data/download_sedata_onlyval.sh
then, run:
python ./code/senet_infer.py
The denoised files will be stored in the folder dataset/valset_noisy_denoised/, with the same name as the corresponding source files in dataset/valset_noisy/.
The dataset can be automatically downloaded and pre-processed (i.e. resampled at 16kHz) by running the script
./data/download_sedata.sh
To download only the testing data, you can run the reduced script:
./data/download_sedata_onlyval.sh
-dataset
- trainset_noisy/ (for the noisy speech training files),
- trainset_clean/ (for the ground truth clean speech training files),
- valset_noisy/ (for the noisy validation files), and
- valset_clean/ (for the noisy validation files).
Once you've downloaded in the script download_data.sh, you can directly train a model using the training dataset by running
python ./code/senet_train.py
The trained model will be stored in the pretrained folder with the names se_model.ckpt.*.
Downloading and pre-processing (i.e., downsampling to 16kHz) the corresponding data can be done by running the script: ./data/download_lossdata.sh
Once the data is downloaded, you can (re-)train a deep feature loss model by running:
python ./code/lossnet_train.py
The loss model is stored in the pretrained folder by default. A custom output directory for loss model can be specified as:
python ./code/lossnet_train.py -o out_folder