FullSubNet

https://fullsubnet.readthedocs.io/en/latest/
Admin
FullSubNet is a full-band and sub-band fusion model for single-channel real-time speech enhancement...
Tokens:2,400
Snippets:32
Trust Score:9.6
Update:5 months ago
Show doc for...
Tokens:
Raw
### Clone Repository

Source: https://fullsubnet.readthedocs.io/en/latest/usage/prerequisites.html

Clone the FullSubNet repository to your local machine. This is the first step before proceeding with environment setup.

```bash
git clone https://github.com/haoxiangsnr/FullSubNet
cd FullSubNet
```

--------------------------------

### Install Pypi Packages

Source: https://fullsubnet.readthedocs.io/en/latest/usage/prerequisites.html

Install Python packages from PyPI, including Cython, librosa, and others. Some packages like 'torch_complex' and 'pesq' might require specific installation methods.

```bash
# you may need to install Cython firstly
pip install Cython
```

```bash
pip install librosa tbb tensorboard joblib matplotlib pesq pystoi tqdm toml torch_complex rich
```

```bash
# install pypesq directly from the GitHub repository
pip install https://github.com/vBaiCai/python-pesq/archive/master.zip
```

--------------------------------

### Install FFmpeg for MP3 Support

Source: https://fullsubnet.readthedocs.io/en/latest/usage/prerequisites.html

Install FFmpeg using Conda if your dataset contains MP3 audio files. This is an optional step required for handling MP3 format.

```bash
# (Optional) if there are "mp3" format audio files in your dataset, you need to install ffmpeg
conda install -c conda-forge ffmpeg
```

--------------------------------

### Install Conda Packages

Source: https://fullsubnet.readthedocs.io/en/latest/usage/prerequisites.html

Install core packages like PyTorch, torchvision, torchaudio, and cudatoolkit using Conda. It's recommended to install conda packages before pip packages.

```bash
conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch
```

```bash
conda install tensorboard joblib matplotlib -c conda-forge
```

--------------------------------

### Visualize Training Logs with TensorBoard

Source: https://fullsubnet.readthedocs.io/en/latest/usage/getting_started.html

Visualize training progress, including loss curves, spectrograms, and audio files, using TensorBoard. Ensure the log directory path is correctly specified.

```bash
tensorboard --logdir ~/Experiments/FullSubNet/train

# specify a port 45454
tensorboard --logdir ~/Experiments/FullSubNet/train --port 45454
```

--------------------------------

### Train FullSubNet Model

Source: https://fullsubnet.readthedocs.io/en/latest/usage/getting_started.html

Use this command to train the FullSubNet model with a default configuration and two GPUs. Ensure you are in the dataset directory.

```bash
# enter a directory named after the dataset, such as dns_interspeech_2020
cd FullSubNet/recipes/dns_interspeech_2020

# Use a default config and two GPUs to train the FullSubNet model
CUDA_VISIBLE_DEVICES=0,1
torchrun --standalone --nnodes=1 --nproc_per_node=2 train.py -C fullsubnet/train.toml
```

--------------------------------

### Train Fullband Baseline Model

Source: https://fullsubnet.readthedocs.io/en/latest/usage/getting_started.html

Train the Fullband baseline model using a default configuration and one GPU. This command is similar to training FullSubNet but uses a different configuration file.

```bash
# Use default config and one GPU to train the Fullband baseline model
CUDA_VISIBLE_DEVICES=0
torchrun --standalone --nnodes=1 --nproc_per_node=1 train.py -C fullband_baseline/train.toml
```

--------------------------------

### Create Conda Environment

Source: https://fullsubnet.readthedocs.io/en/latest/usage/prerequisites.html

Create and activate a new Conda environment named 'FullSubNet' with Python 3.10. Optionally, remove an existing environment with the same name first.

```bash
# (Optional) if you already have an environment with the same name, you may remove it first
conda env remove --name FullSubNet

# create a new environment
conda create --name FullSubNet python=3.10
conda activate FullSubNet
```

--------------------------------

### Configure Inference Dataset

Source: https://fullsubnet.readthedocs.io/en/latest/usage/getting_started.html

Configure the dataset directory and sample rate for speech enhancement inference. Update `dataset_dir_list` with your noisy speech data paths.

```toml
[dataset.args]
dataset_dir_list = [
    "/path/to/your/dataset_1/",
    "/path/to/your/dataset_2/",
    "others..."
]
sr = 16000
```

--------------------------------

### Calculate Speech Enhancement Metrics

Source: https://fullsubnet.readthedocs.io/en/latest/usage/getting_started.html

Calculate performance metrics such as SI_SDR, STOI, and PESQ for the enhanced speech. Provide paths to reference and enhanced audio, and specify the metrics to compute.

```bash
# Switch path
cd FullSubNet

# DNS-INTERSPEECH-2020
python tools/calculate_metrics.py \
  -R /path/to/reference/ \
  -E /path/to/enhancement/ \
  -M SI_SDR,STOI,WB_PESQ,NB_PESQ \
  -S DNS_1
```

--------------------------------

### Resume Training Experiment

Source: https://fullsubnet.readthedocs.io/en/latest/usage/getting_started.html

Resume a previously interrupted training experiment using the '-R' parameter. This allows you to continue training from the last saved checkpoint.

```bash
# Resume the experiment using "-R" parameter
CUDA_VISIBLE_DEVICES=0,1
torchrun --standalone --nnodes=1 --nproc_per_node=2 train.py -C fullband_baseline/train.toml -R
```

--------------------------------

### Run Inference for Speech Enhancement

Source: https://fullsubnet.readthedocs.io/en/latest/usage/getting_started.html

Perform speech enhancement using a trained model. Specify the inference configuration file, the path to the best model checkpoint, and the output directory.

```bash
cd FullSubNet/recipes/dns_interspeech_2020

# One GPU is used by default
python inference.py \
  -C fullsubnet/inference.toml \
  -M /path/to/your/checkpoint_dir/best_model.tar \
  -O /path/to/your/enhancement/dir
```

--------------------------------

### Angular Commit Style

Source: https://fullsubnet.readthedocs.io/en/latest/reference/contributing.html

Follow this commit message structure for consistency and automated version bumping. Ensure a type and summary are always provided.

```git
<type>(optional scope): short summary in present tense

(optional body: explains motivation for the change)

(optional footer: note BREAKING CHANGES here, and issues to be closed)

```
Tokens:
Raw