Wav2CLIP: Learning Robust Audio Representations From CLIP

TL;DR

Wav2CLIP is a novel audio representation learning method that leverages CLIP to create a shared multimodal embedding space, enabling efficient, zero-shot audio tasks and cross-modal applications with less data and training effort.

Contribution

It introduces Wav2CLIP, a new approach that distills audio representations from CLIP, outperforming existing methods and enabling multimodal applications without extensive visual model training.

Findings

Wav2CLIP outperforms several existing pre-trained audio models.

It achieves competitive performance with only ~10% of the data used by supervised models.

Wav2CLIP enables image generation from audio embeddings.

Abstract

We propose Wav2CLIP, a robust audio representation learning method by distilling from Contrastive Language-Image Pre-training (CLIP). We systematically evaluate Wav2CLIP on a variety of audio tasks including classification, retrieval, and generation, and show that Wav2CLIP can outperform several publicly available pre-trained audio representation algorithms. Wav2CLIP projects audio into a shared embedding space with images and text, which enables multimodal applications such as zero-shot classification, and cross-modal retrieval. Furthermore, Wav2CLIP needs just ~10% of the data to achieve competitive performance on downstream tasks compared with fully supervised models, and is more efficient to pre-train than competing methods as it does not require learning a visual model in concert with an auditory model. Finally, we demonstrate image generation from Wav2CLIP as qualitative assessment of the shared embedding space. Our code and model weights are open sourced and made available for further applications.