### Install Fairseq

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/README.md

Install the fairseq library by navigating to its directory and using pip with an editable install.

```Shell
cd tools/fairseq/
pip install --editable .
```

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

### Install Face Detection

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/README.md

Install the face_detection library using pip with an editable install.

```Shell
cd ../face_detection/
pip install --editable .
```

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

### Install Face Alignment

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/README.md

Install the face_alignment library using pip with an editable install.

```Shell
cd ../face_alignment/
pip install --editable .
```

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

### Install Other Dependencies

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/README.md

Install remaining project dependencies listed in the requirements file.

```Shell
cd ../../
pip install -r reqs.txt
```

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

### Install PyTorch and Related Libraries

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/README.md

Install specific versions of PyTorch, torchaudio, torchvision, and other related libraries using conda.

```Shell
conda install pytorch-lightning==1.9.3 pytorch==2.0.1 torchaudio==2.0.2 torchvision==0.15.2 torchmetrics==0.11.2 pytorch-cuda==11.8 cudatoolkit==11.8 -c pytorch -c nvidia -y
```

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

### Configure Training for Full Dataset (Stage 2)

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/README.md

Initialize the model with the checkpoint from Stage 1 and train on the full dataset. This configuration uses the checkpoint from the short utterance training.

```Shell
# Example configuration setting:
# Use checkpoint from Stage 1
# Train with full dataset
```

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

### Run Testing with main.py

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/README.md

Use main.py for testing. Ensure the correct config filename is specified within the script.

```python
import main

# Specify the config filename here
main.recognize("your_config_filename.yaml")
```

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

### Running Data Preparation Scripts

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/preparation/README.md

These Python scripts are used to prepare the dataset. `prepare_filescp.py` generates a list file, and `detect_landmark_list.py` extracts lip videos based on this list. Modify the bash variables to match your paths and dataset configuration.

```Shell
python prepare_filescp.py \
    --src $DOWNLOAD_DATA_PATH \
    --dst $TARGET_DATA_PATH \
    --dataset $DATASET_NAME \
    --split $SPLIT

python detect_landmark_list.py --list $DATASET_NAME-$SPLIT.scp --rank 0 --shard 1
```

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

### Clone Repository and Initialize Submodules

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/README.md

Clone the repository, navigate into it, and initialize/update git submodules. This step also includes pulling large files using git-lfs.

```Shell
git clone git@github.com:MKT-Dataoceanai/CNVSRC2023Baseline.git
cd CNVSRC2023Baseline
git submodule init
git submodule update
cd tools/face_detection
git lfs pull
cd ../../
```

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

### Run Training from Pre-trained Model

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/README.md

Execute the main training script using a pre-trained model configuration. Ensure 'ckpt_path' and 'remove_ctc' are set appropriately in the YAML file.

```Shell
python main.py
```

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

### Train VSR/ASR Model on Custom Dataset

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/INSTRUCTION.md

Use this command to initiate training on a custom dataset. Set `data/dataset` to `cstm` to load the custom dataset. Ensure all placeholder values like `[exp_dir]`, `[exp_name]`, `[modality]`, `[ckpt_path]`, and `[num_nodes]` are replaced with your specific configuration.

```Shell
python main.py exp_dir=[exp_dir] \
               exp_name=[exp_name] \
               data.modality=[modality] \
               ckpt_path=[ckpt_path] \
               data/dataset=[dataset] \
               trainer.num_nodes=[num_nodes]
```

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

### Configure Training for Short Utterances (Stage 1)

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/README.md

Configure training to use only short utterances (<= 4 seconds/100 frames) by setting 'data.max_length=100' in the YAML file. This stage also sets the learning rate.

```Shell
# Example configuration setting:
data.max_length=100
optimizer.lr=0.0002
```

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

### Run Testing with predict.py

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/README.md

Use predict.py for testing to observe real-time output. Ensure the correct config filename is specified within the script.

```python
import predict

# Specify the config filename here
predict.recognize("your_config_filename.yaml")
```

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

### Save Pre-processed Audio-Visual Pairings and Text

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/INSTRUCTION.md

Use this code to save pre-processed video, audio, and text data for a custom dataset. Ensure the AVSRDataLoader and save_vid_aud_txt functions are imported.

```Python
from preparation.data.data_module import AVSRDataLoader
from preparation.utils import save_vid_aud_txt

# Initialize video and audio data loaders
video_loader = AVSRDataLoader(modality="video", detector="retinaface", convert_gray=False)
audio_loader = AVSRDataLoader(modality="audio")

# Specify the file path to the data
data_path = 'data_filename'

# Load video and audio data from the same data file
video_data = video_loader.load_data(data_path)
audio_data = audio_loader.load_data(data_path)

# Load text
text = ...

# Define output paths for the processed video, audio, and text data
output_video_path = 'cstm/cstm_video_seg24s/test_file_1.mp4'
output_audio_path = 'cstm/cstm_video_seg24s/test_file_1.wav'
output_text_path = 'cstm/cstm_text_seg24s/test_file_1.txt'

# Save the loaded video, audio, and associated text data
save_vid_aud_txt(output_video_path, output_audio_path, output_text_path, video_data, audio_data, text, video_fps=25, audio_sample_rate=16000)
```

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

### Create and Activate Conda Environment

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/README.md

Create a new conda environment named 'cnvsrc' with Python 3.10.11 and activate it.

```Shell
conda create -y -n cnvsrc python==3.10.11
conda activate cnvsrc
```

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

### Decompressing CNVSRC2023 Dataset Files

Source: https://github.com/dataoceanai/cnvsrc2023baseline/blob/master/preparation/README.md

Execute these shell commands to decompress all downloaded tar.gz files for both the development and evaluation sets of the CNVSRC2023 dataset. Ensure you are in the correct directory before running.

```Shell
# cd CNVSRC/CNVSRC2023/
# cd cncvs/news/
# tar -xzvf news_part01.tar.gz
# tar -xzvf news_part02.tar.gz
# cd ../speech/
# tar -xzvf speech_part01.tar.gz
# tar -xzvf speech_part02.tar.gz
# tar -xzvf speech_part03.tar.gz
# tar -xzvf speech_part04.tar.gz
# tar -xzvf speech_part05.tar.gz
# tar -xzvf speech_part06.tar.gz
# tar -xzvf speech_part07.tar.gz
# tar -xzvf speech_part08.tar.gz
# tar -xzvf speech_part09.tar.gz
# tar -xzvf speech_part10.tar.gz
# tar -xzvf speech_part11.tar.gz
# tar -xzvf speech_part12.tar.gz
# cd ../../
# tar -xzvf cnvsrc-multi-dev.tar.gz
# tar -xzvf cnvsrc-single-dev.tar.gz
```

```Shell
# cd CNVSRC/CNVSRC2023/
# tar -xzvf cnvsrc-multi-eval.tar.gz
# tar -xzvf cnvsrc-single-eval.tar.gz
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.