CLAP

CLAP (Contrastive Language-Audio Pretraining) is a neural network model that learns acoustic concepts from natural language supervision. It achieves SoTA in “Zero-Shot” classification, Audio-Text & Text-Audio Retrieval, and in some datasets when finetuned.

Citation

https://arxiv.org/pdf/2206.04769.pdf

@article{elizalde2022clap,
  title={Clap: Learning audio concepts from natural language supervision},
  author={Elizalde, Benjamin and Deshmukh, Soham and Ismail, Mahmoud Al and Wang, Huaming},
  journal={arXiv preprint arXiv:2206.04769},
  year={2022}
}

Request CLAP weights:

https://forms.office.com/r/ULb4k9GL1F

Usage

Load model

from CLAP_API import CLAP 

clap_model = CLAP("<PATH TO WEIGHTS>", use_cuda=False)

Extract text embeddings


text_embeddings = clap_model.get_text_embeddings(class_labels: List[str])

Extract audio embeddings


audio_embeddings = clap_model.get_audio_embeddings(file_paths: List[str])

Compute similarity

# For using the below function, DO NOT normalize the text and audio embeddings
sim = clap_model.compute_similarity(audio_embeddings, text_embeddings)

Zero-Shot Classification of ESC50 dataset

from CLAP_API import CLAP
from esc50 import ESC50
import torch.nn.functional as F
import numpy as np
from tqdm import tqdm
from sklearn.metrics import accuracy_score

# Load CLAP
weights_path = # Add weight path here
clap_model = CLAP(weights_path, use_cuda=False)

# Load dataset
dataset = ESC50(root='path/ESC-50-master', download=False)
prompt = 'this is a sound of '
Y = [prompt + x for x in dataset.classes]

# Computing text embeddings
text_embeddings = clap_model.get_text_embeddings(Y)

# Computing audio embeddings
y_preds, y_labels = [], []
for i in tqdm(range(len(dataset))):
    x, _, one_hot_target = dataset.__getitem__(i)

    audio_embeddings = clap_model.get_audio_embeddings([x], resample=True)
    similarity = clap_model.compute_similarity(audio_embeddings, text_embeddings)
    y_pred = F.softmax(similarity.detach().cpu(), dim=1).numpy()
    y_preds.append(y_pred)
    y_labels.append(one_hot_target.detach().cpu().numpy())

y_labels, y_preds = np.concatenate(y_labels, axis=0), np.concatenate(y_preds, axis=0)
acc = accuracy_score(np.argmax(y_labels, axis=1), np.argmax(y_preds, axis=1))
print('ESC50 Accuracy {}'.format(acc))

The output:

ESC50 Accuracy: 82.6%

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.

4.0 KiB Raw Blame History