### Install Training Dependencies

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Installs necessary packages for training, including support for PEP 660. Ensure pip is up-to-date before running.

```bash
pip install --upgrade pip  # enable PEP 660 support
pip install -e .[train,optional]
```

```zsh
pip install -e '.[train,optional]'
```

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

### Install UTMOSv2 Dependencies

Source: https://github.com/sarulab-speech/utmosv2/blob/main/README.md

Install the project dependencies using uv or pip. The '[optional]' flag installs additional dependencies required for certain features.

```bash
uv sync --extra optional
```

```bash
pip install --upgrade pip  # enable PEP 660 support
pip install -e .[optional] # install with optional dependencies
```

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

### Install UTMOSv2

Source: https://github.com/sarulab-speech/utmosv2/blob/main/quickstart.ipynb

Installs the UTMOSv2 library from GitHub. Ensure Git LFS is configured to skip smudging if needed.

```python
!GIT_LFS_SKIP_SMUDGE=1 pip install git+https://github.com/sarulab-speech/UTMOSv2.git
```

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

### Install Training Dependencies

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/training.md

Installs the necessary dependencies for training UTMOSv2, including support for PEP 660. Ensure pip is up-to-date before running.

```bash
pip install --upgrade pip  # enable PEP 660 support
pip install -e .[train,optional]
```

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

### Install UTMOSv2 Library

Source: https://github.com/sarulab-speech/utmosv2/blob/main/README.md

Install the UTMOSv2 library from GitHub using uv or pip. This is the first step before using the library in your Python code.

```bash
uv add git+https://github.com/sarulab-speech/UTMOSv2.git
# If you're using pip:
# pip install git+https://github.com/sarulab-speech/UTMOSv2.git
```

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

### Install Training Dependencies (zsh)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/training.md

Installs training dependencies for zsh users, requiring escaped square brackets in the pip command.

```zsh
pip install -e '.[train,optional]'
```

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

### Train UTMOSv2 with Weights & Biases

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/training.md

Command to enable Weights & Biases (wandb) for experiment tracking during training. Requires API key setup.

```bash
python train.py --config spec_only --data_config data_config.json --wandb
```

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

### Install Optional Dependencies for Inference Scripts

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/inference.md

Install additional dependencies required for using the `inference.py` scripts. Ensure pip is up-to-date for PEP 660 support.

```bash
pip install --upgrade pip  # enable PEP 660 support
pip install -e .[optional]
```

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

### Install Optional Dependencies with Zsh Escaping

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/inference.md

If using zsh, escape the square brackets when installing optional dependencies for the inference scripts.

```zsh
pip install -e '.[optional]'
```

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

### Data Configuration JSON Structure

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/training.md

Example JSON structure for configuring custom training data. Specifies dataset name, directory, and MOS list file.

```json
{
  "data": [
    {
      "name": "dataset1",
      "dir": "/path/to/your/dataset1",
      "mos_list": "/path/to/your/moslist1.txt"
    },
    {
      "name": "dataset2",
      "dir": "/path/to/your/dataset2",
      "mos_list": "/path/to/your/moslist2.txt"
    }
    // Add more data entries as needed
  ]
}
```

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

### Train UTMOSv2 Paper Experiments (SSL Stage 1)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Use this command to train the first stage of the SSL-only model for UTMOSv2.

```bash
python train.py --reproduce --config ssl_only_stage1
```

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

### Clone UTMOSv2 Repository

Source: https://github.com/sarulab-speech/utmosv2/blob/main/README.md

Clone the UTMOSv2 repository and navigate to its directory. This is the first step to set up the project locally.

```bash
git clone https://github.com/sarulab-speech/UTMOSv2.git
cd UTMOSv2
```

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

### Run Inference Script with Configuration

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/inference.md

Use this command to run the inference script with a specified configuration file, input directory, and output path. By default, it uses the fusion_stage3 model.

```bash
python inference.py --config configuration_file_name --input_dir /path/to/wav/dir/ --out_path /path/to/output/file.csv
```

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

### Train UTMOSv2 with Custom Data

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/training.md

Command to initiate UTMOSv2 training using a specified data configuration file.

```bash
python train.py --config spec_only --data_config data_config.json
```

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

### Fine-tune UTMOSv2 from Pre-trained Weights

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/training.md

Command to continue training from existing weights by specifying the --weight option. Useful for further training or fine-tuning.

```bash
python train.py --config spec_only --data_config data_config.json --weight /path/to/your/weights.pth
```

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

### Specify Configuration File for Predictions in Python

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/inference.md

Create a UTMOSv2 model by providing a configuration file name to `utmosv2.create_model` in Python.

```python
model = utmosv2.create_model(config="configuration_file_name")
mos = model.predict(input_dir="/path/to/wav/dir/")
```

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

### Train UTMOSv2 Paper Experiments (SSL Stage 2)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Use this command to train the second stage of the SSL-only model, requiring the first stage weights.

```bash
python train.py --reproduce --config ssl_only_stage2 --weight ssl_only_stage1
```

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

### Train UTMOSv2 Paper Experiments (w/o SSL)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Use this command to train the UTMOSv2 system without SSL components, focusing on spectral information.

```bash
python train.py --reproduce --config spec_only
```

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

### Train Competition Submission (SSL Stage 1)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Use this command to train the first stage of the SSL-only model for the competition submission.

```bash
python train.py --reproduce --config c_ssl_only_stage1
```

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

### Train Competition Submission (SSL Stage 2)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Use this command to train the second stage of the SSL-only model for the competition submission, requiring the first stage weights.

```bash
python train.py --reproduce --config c_ssl_only_stage2 --weight ssl_only_stage1
```

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

### Train Competition Submission (Spectral Stage 1)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Use this command to train the first stage of the spectral-only model for the competition submission.

```bash
python train.py --reproduce --config c_spec_only_stage1
```

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

### Train UTMOSv2 Paper Experiments (Fusion w/o Stage 1 & 2)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Use this command to train the fusion model excluding both stage 1 and stage 2.

```bash
python train.py --reproduce --config fusion_wo_stage1and2
```

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

### Train UTMOSv2 Paper Experiments (Fusion w/o Stage 2)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Use this command to train the fusion model excluding stage 2.

```bash
python train.py --reproduce --config fusion_wo_stage2
```

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

### Predict MOS for WAV Files in a Directory

Source: https://github.com/sarulab-speech/utmosv2/blob/main/quickstart.ipynb

Creates a pre-trained UTMOSv2 model and predicts the MOS for all WAV files within a specified directory.

```python
model = utmosv2.create_model(pretrained=True)
mos = model.predict(input_dir="/path/to/wav/dir/")
print(mos)
```

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

### Train Competition Submission (Spectral Stage 2)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Use this command to train the second stage of the spectral-only model for the competition submission, requiring the first stage weights.

```bash
python train.py --reproduce --config c_spec_only_stage2 --weight c_spec_only_stage1
```

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

### Train UTMOSv2 Paper Experiments (Fusion Stage 2)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Use this command to train the second stage of the fusion model for UTMOSv2.

```bash
python train.py --reproduce --config fusion_stage2
```

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

### Train Competition Submission (Fusion Stage 2)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Use this command to train the second stage of the fusion model for the competition submission.

```bash
python train.py --reproduce --config c_fusion_stage2
```

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

### Train Competition Submission (Fusion Stage 3)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Use this command to train the third stage of the fusion model for the competition submission, requiring the second stage weights.

```bash
python train.py --reproduce --config c_fusion_stage3 --weight c_fusion_stage2
```

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

### Create Model and Predict in Python

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/inference.md

Integrate UTMOSv2 into your Python code by creating a model instance with a specific checkpoint path and then using the predict method.

```python
model = utmosv2.create_model(checkpoint_path="/path/to/your/weight.pth")
MOS = model.predict(input_dir="/path/to/wav/dir/")
```

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

### Train UTMOSv2 Paper Experiments (Fusion Stage 3)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Use this command to train the third stage of the fusion model, requiring the second stage weights.

```bash
python train.py --reproduce --config fusion_stage3 --weight fusion_stage2
```

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

### Inference with Trained Weights

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Run inference using trained weights. Specify the configuration name, input directory, validation list path, and number of repetitions.

```bash
python inference.py --reproduce --config fusion_stage3 --input_dir data/main/DATA --val_list_path voicemos2024-track1-eval-phase/DATA/sets/eval_list.txt --fold -1 --num_repetitions 5
```

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

### Predict MOS for a Subset of Files in Python using File Path

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/inference.md

In Python, use the `predict` method with `val_list_path` to specify a text file containing the list of WAV files for prediction.

```python
mos = model.predict(input_dir="/path/to/wav/dir/", val_list_path="/path/to/your/val/list.txt")
```

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

### Train UTMOSv2 Paper Experiments (w/o BVCC Datasets)

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/reproduction.md

Commands to train UTMOSv2 experiments excluding BVCC datasets. Similar replacements can be made for 'bc', 'somos', and 'sarulab'.

```bash
python train.py --reproduce --config spec_only_wo_bvcc
```

```bash
python train.py --reproduce --config ssl_only_stage1_wo_bvcc
```

```bash
python train.py --reproduce --config ssl_only_stage2_wo_bvcc --weight ssl_only_stage1_wo_bvcc
```

```bash
python train.py --reproduce --config fusion_stage2_wo_bvcc
```

```bash
python train.py --reproduce --config fusion_stage3_wo_bvcc --weight fusion_stage2_wo_bvcc
```

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

### Run Inference Script with Custom Weights

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/inference.md

Execute the inference script using custom weights by specifying the --weight option along with the input and output paths.

```bash
python inference.py --input_dir /path/to/wav/dir/ --out_path /path/to/output/file.csv --weight /path/to/your/weight.pth
```

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

### Predict MOS for a Single WAV File

Source: https://github.com/sarulab-speech/utmosv2/blob/main/quickstart.ipynb

Creates a pre-trained UTMOSv2 model and predicts the MOS for a single audio file specified by its path.

```python
model = utmosv2.create_model(pretrained=True)
mos = model.predict(input_path="/path/to/wav/file.wav")
print(mos)
```

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

### Predict MOS for All WAV Files in a Directory

Source: https://github.com/sarulab-speech/utmosv2/blob/main/README.md

Use the inference script to predict the MOS for all WAV files within a specified directory. Provide the input directory path and the output CSV file path.

```python
python inference.py --input_dir /path/to/wav/dir/ --out_path /path/to/output/file.csv
```

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

### Predict MOS for a Subset of Files using Inference Script

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/inference.md

Use the `inference.py` script with `val_list_path` to predict MOS scores for a specified subset of WAV files.

```bash
python inference.py --input_dir /path/to/wav/dir/ --out_path /path/to/output/file.csv --val_list_path /path/to/your/val/list.txt
```

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

### Import UTMOSv2 Library

Source: https://github.com/sarulab-speech/utmosv2/blob/main/quickstart.ipynb

Imports the necessary utmosv2 library for use in your Python script.

```python
import utmosv2
```

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

### Run Inference with Different Weights per Fold

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/inference.md

Iterate through folds and run the inference script using distinct weights for each fold by specifying the --weight and --fold arguments.

```bash
for i in {0..5}; do
    python inference.py --input_path /path/to/wav/file.wav --out_path /path/to/output/file.csv --weight /path/to/your/weight_fold${i}.pth --fold $i
done
```

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

### Predict MOS for a Subset of Files in Python using Direct List

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/inference.md

Provide a Python list of utterance IDs directly to the `predict` method to process only specific files.

```python
mos = model.predict(
    input_dir="/path/to/wav/dir/",
    val_list=["sys00691-utt0682e32", "sys00691-utt31fd854", "sys00691-utt33a4826", ...]
)
```

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

### Specify Fold and Repetitions for Accurate Predictions in Python

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/inference.md

Create a UTMOSv2 model specifying a fold and use the `predict` method with `num_repetitions` for more accurate predictions in Python.

```python
model = utmosv2.create_model(fold=2)
mos = model.predict(input_dir="/path/to/wav/dir/", num_repetitions=5)
```

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

### Predict MOS with 'somos' Data Domain using Inference Script

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/inference.md

Execute the `inference.py` script to predict MOS scores, setting the data domain to 'somos'.

```bash
python inference.py --input_dir /path/to/wav/dir/ --out_path /path/to/output/file.csv --predict_dataset somos
```

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

### Specify Fold and Repetitions for Accurate Predictions using Inference Script

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/inference.md

Run the `inference.py` script, specifying the `--fold` and `--num_repetitions` arguments for enhanced prediction accuracy.

```bash
python inference.py --input_dir /path/to/wav/dir/ --out_path /path/to/output/file.csv --fold 2 --num_repetitions 5
```

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

### Predict MOS with 'somos' Data Domain in Python

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/inference.md

Use the `predict` method in Python to make MOS predictions, specifying the 'somos' data domain.

```python
mos = model.predict(input_dir="/path/to/wav/dir/", predict_dataset="somos")
```

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

### Predict MOS from Tensor or Array

Source: https://github.com/sarulab-speech/utmosv2/blob/main/README.md

Use the UTMOSv2 library to predict MOS scores from audio data already loaded in memory as a PyTorch tensor or NumPy array. Ensure the data has the correct shape and sampling rate.

```python
import utmosv2
model = utmosv2.create_model(pretrained=True)
# data: torch.Tensor or np.ndarray with shape (batch_size, sequence_length) or (sequence_length,)
# sr: Sampling rate of the input audio data. If not provided, it defaults to 16000 Hz.
mos = model.predict(data=data, sr=16000) # Returns a torch.Tensor or np.ndarray with shape (batch_size,) or (1,)
```

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

### MOS List File Format

Source: https://github.com/sarulab-speech/utmosv2/blob/main/docs/training.md

Format for the MOS list file, mapping audio file identifiers to their corresponding MOS values.

```text
sys64e2f-utt491a78a,2.375
sys64e2f-utt8485f83,3.625
sys7ab3c-utt1417b69,4.0
...
```

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

### Predict MOS for a Single WAV File

Source: https://github.com/sarulab-speech/utmosv2/blob/main/README.md

Use the inference script to predict the Mean Opinion Score (MOS) for a single WAV file. Specify the input WAV file path and the output CSV file path.

```python
python inference.py --input_path /path/to/wav/file.wav --out_path /path/to/output/file.csv
```

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

### UTMOSv2 Citation

Source: https://github.com/sarulab-speech/utmosv2/blob/main/README.md

BibTeX entry for citing the UTMOSv2 paper in academic publications. Use this when referencing the project in research.

```bibtex
@inproceedings{baba2024utmosv2,
  title     = {The T05 System for The {V}oice{MOS} {C}hallenge 2024: Transfer Learning from Deep Image Classifier to Naturalness {MOS} Prediction of High-Quality Synthetic Speech},
  author    = {Baba, Kaito and Nakata, Wataru and Saito, Yuki and Saruwatari, Hiroshi},
  booktitle = {IEEE Spoken Language Technology Workshop (SLT)},
  year      = {2024},
  pages     = {818--824},
  doi       = {10.1109/SLT61566.2024.10832315},
}
```

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