Open-Unmix

https://sigsep.github.io/open-unmix/
Admin
Open-Unmix is a deep neural network reference implementation for music source separation that...
Tokens:4,166
Snippets:78
Trust Score:9.7
Update:6 months ago
Show doc for...
Tokens:
Raw
### Install open-unmix

Source: https://sigsep.github.io/open-unmix

Install the open-unmix package from PyPI. For extended audio format support, also install stempeg.

```bash
pip install openunmix

```

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

### Install Norbert

Source: https://sigsep.github.io/open-unmix/norbert.html

Use pip to install the package.

```bash
pip install norbert
```

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

### Install museval package

Source: https://sigsep.github.io/open-unmix/museval.html

Install the museval parsing package using pip.

```bash
pip install museval

```

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

### Load Pre-trained Model (Installed Package)

Source: https://sigsep.github.io/open-unmix

Load a pre-trained model, such as 'umxhq', directly from the installed open-unmix package in Python.

```python
separator = openunmix.umxhq(...)

```

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

### Docker Installation and Usage

Source: https://sigsep.github.io/open-unmix

Use the provided Docker container to perform audio separation on a local track. Mount your music directory to /data within the container.

```bash
docker run -v ~/Music/:/data -it faroit/open-unmix-pytorch umx "/data/track1.wav" --outdir /data/track1

```

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

### Load Pre-trained Model (torch.hub)

Source: https://sigsep.github.io/open-unmix

Load a pre-trained model using torch.hub without installing the open-unmix package. Ensure PyTorch is installed.

```python
separator = torch.hub.load('sigsep/open-unmix-pytorch', 'umxhq', device=device)

```

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

### Evaluate MUSDB using museval CLI

Source: https://sigsep.github.io/open-unmix/museval.html

Run the museval command line tool to evaluate estimates against the MUSDB dataset.

```bash
museval -p --musdb path/to/musdb -o path/to/output_dir path/to/estimate_dir
```

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

### Pull Docker evaluation image

Source: https://sigsep.github.io/open-unmix/museval.html

Download the precompiled sigsep-mus-eval Docker image from Docker Hub.

```bash
docker pull faroit/sigsep-mus-eval
```

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

### Run evaluation in Docker

Source: https://sigsep.github.io/open-unmix/museval.html

Execute the evaluation inside a Docker container by mounting absolute paths for estimates, MUSDB, and output directories.

```bash
docker run --rm -v estimatesdir:/est -v musdbdir:/mus -v outputdir:/out faroit/sigsep-mus-eval --musdb /mus -o /out /est
```

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

### Compare Multiple Methods with MethodStore

Source: https://sigsep.github.io/open-unmix/museval.html

Use museval.MethodStore to compare results from multiple evaluation runs. You can add existing EvalStore objects or load scores from SiSEC MUS 2018.

```python
methods = museval.MethodStore()
methods.add_evalstore(results, name="XZY")


```

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

### Command-line Audio Separation

Source: https://sigsep.github.io/open-unmix

Separate audio files using a specified pre-trained model from the command line. Supports WAV, FLAC, and OGG formats.

```bash
umx input_file.wav --model umxl

```

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

### Evaluate MUSDB18 tracks on-the-fly

Source: https://sigsep.github.io/open-unmix/museval.html

This function separates a mixture into vocals and accompaniment tracks and evaluates them using museval. Ensure the output_dir is set to save evaluation results.

```python
import musdb
import museval

def estimate_and_evaluate(track):
    # assume mix as estimates
    estimates = {
        'vocals': track.audio,
        'accompaniment': track.audio
    }

    # Evaluate using museval
    scores = museval.eval_mus_track(
        track, estimates, output_dir="path/to/json"
    )

    # print nicely formatted and aggregated scores
    print(scores)

mus = musdb.DB()
for track in mus:
    estimate_and_evaluate(track)

```

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

### Aggregate and Analyze Scores with EvalStore

Source: https://sigsep.github.io/open-unmix/museval.html

Aggregate scores for multiple tracks into a pandas DataFrame using museval.EvalStore for easier analysis. Scores can be aggregated using median for frames and tracks.

```python
results = museval.EvalStore(frames_agg='median', tracks_agg='median')
for track in tracks:
    # ...
    results.add_track(museval.eval_mus_track(track, estimates))

```

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

### Evaluate MUSDB18 tracks later

Source: https://sigsep.github.io/open-unmix/museval.html

Evaluate pre-computed estimates from a directory using museval.eval_mus_dir. This function accepts parameters from musdb.run() for convenience, such as enabling multiprocessing.

```python
import musdb
import museval

# initiate musdb
mus = musdb.DB()

# evaluate an existing estimate folder with wav files
museval.eval_mus_dir(
    dataset=mus,  # instance of musdb
    estimates_dir=...,  # path to estimate folder
    output_dir=...,  # set a folder to write eval json files
    subsets="test",
    is_wav=False
)

```

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

### Unified Separation Function

Source: https://sigsep.github.io/open-unmix

Use the 'separate.predict' function for a streamlined process of model loading, preprocessing, and audio separation.

```python
from openunmix import separate
estimates = separate.predict(audio, ...)

```

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

### Perform Audio Separation

Source: https://sigsep.github.io/open-unmix

Separate audio data using a loaded separator object. The audio must be a torch.Tensor with the correct shape and sampling rate.

```python
estimates = separator(audio, ...)

```

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

### Cite SiSEC 2018

Source: https://sigsep.github.io/open-unmix/museval.html

BibTeX citation for the 2018 Signal Separation Evaluation Campaign.

```bibtex
@InProceedings{SiSEC18,
  author="St{\"o}ter, Fabian-Robert and Liutkus, Antoine and Ito, Nobutaka",
  title="The 2018 Signal Separation Evaluation Campaign",
  booktitle="Latent Variable Analysis and Signal Separation:
  14th International Conference, LVA/ICA 2018, Surrey, UK",
  year="2018",
  pages="293--305"
}
```

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

### Perform Multichannel Wiener Filtering

Source: https://sigsep.github.io/open-unmix/norbert.html

Apply the multichannel Wiener filter to a complex spectrogram and target estimates.

```python
X = stft(audio)
V = model(X)
Y = norbert.wiener(V, X)
estimate = istft(Y)
```
Tokens:
Raw