Official implementation of the paper WAV2CLIP: LEARNING ROBUST AUDIO REPRESENTATIONS FROM CLIP

Last update: Dec 13, 2022

Related tags

Deep Learning lyrebird-wav2clip

Overview

Wav2CLIP

🚧 WIP 🚧

Official implementation of the paper WAV2CLIP: LEARNING ROBUST AUDIO REPRESENTATIONS FROM CLIP 📄 🔗

Ho-Hsiang Wu, Prem Seetharaman, Kundan Kumar, Juan Pablo Bello

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.

Installation

pip install wav2clip

Usage

Clip-Level Embeddings

import wav2clip

model = wav2clip.get_model()
embeddings = wav2clip.embed_audio(audio, model)

Frame-Level Embeddings

import wav2clip

model = wav2clip.get_model(frame_length=16000, hop_length=16000)
embeddings = wav2clip.embed_audio(audio, model)

Comments

request of projection layer weight

Hi @hohsiangwu , Thanks for great work! Request pre-trained weights of image_transform (MLP layer) for audio-image-language joint embedding space.

Currently, only audio encoders seem to exist in the get_model function. Is there any big problem if I use CLIP embedding (text or image) without projection layer?

opened by SeungHeonDoh 2
Initial checkin for accessing pre-trained model via pip install

I am considering using the release feature of GitHub to host model weights, once the url is added to MODEL_WEIGHTS_URL, and the repository is made public, we should be able to model = torch.hub.load('descriptinc/lyrebird-wav2clip', 'wav2clip', pretrained=True)

opened by hohsiangwu 1
Adding VQGAN-CLIP with modification to generate audio
Adding a working snapshot of original generate.py from https://github.com/nerdyrodent/VQGAN-CLIP/

Modify to add audio related params and functions

Add scripts to generate image and video with options for conditioning and interpolation
opened by hohsiangwu 0
Supervised scenario no transform

In the supervise scenario in the __init__.py the transform flag is not set to True, so the model doesn't contain the MLP layer after training. I'm wondering how you train the MLP layer when using as pretrained.

opened by alirezadir 0
Integrated into VQGAN+CLIP 3D Zooming notebook

Dear researchers,

I integrated Wav2CLIP into a VQGAN+CLIP animation notebook.

It is available on colab here: https://colab.research.google.com/github/pollinations/hive/blob/main/notebooks/2%20Text-To-Video/1%20CLIP-Guided%20VQGAN%203D%20Turbo%20Zoom.ipynb

I'm part of a team creating an open-source generative art platform called Pollinations.AI. It's also possible to use through our frontend if you are interested. https://pollinations.ai/p/QmT7yt67DF3GF4wd2vyw6bAgN3QZx7Xpnoyx98YWEsEuV7/create

Here is an example output: https://user-images.githubusercontent.com/5099901/168467451-f633468d-e596-48f5-8c2c-2dc54648ead3.mp4

opened by voodoohop 0
The details concerning loading raw audio files

Hi !

I haved imported the wave2clip as a package, however when testing, the inputs for the model to extract features are not original audio files. Thus can you provided the details to load the audio files to processed data for the model?

opened by jinx2018 0
torch version

Hi, thanks for sharing the wonderful work! I encountered some issues during pip installing it, so may I ask what is the torch version you used? I cannot find the requirement of this project. Thanks!

opened by annahung31 0
Error when importing after fresh installation on colab

What CUDA and Python versions have you tested the pip package in? After installation on a fresh collab I receive the following error:

OSError Traceback (most recent call last) in () ----> 1 import wav2clip

7 frames /usr/local/lib/python3.7/dist-packages/wav2clip/init.py in () 2 import torch 3 ----> 4 from .model.encoder import ResNetExtractor 5 6

/usr/local/lib/python3.7/dist-packages/wav2clip/model/encoder.py in () 4 from torch import nn 5 ----> 6 from .resnet import BasicBlock 7 from .resnet import ResNet 8

/usr/local/lib/python3.7/dist-packages/wav2clip/model/resnet.py in () 3 import torch.nn as nn 4 import torch.nn.functional as F ----> 5 import torchaudio 6 7

/usr/local/lib/python3.7/dist-packages/torchaudio/init.py in () ----> 1 from torchaudio import _extension # noqa: F401 2 from torchaudio import ( 3 compliance, 4 datasets, 5 functional,

/usr/local/lib/python3.7/dist-packages/torchaudio/_extension.py in () 25 26 ---> 27 _init_extension()

/usr/local/lib/python3.7/dist-packages/torchaudio/_extension.py in _init_extension() 19 # which depends on libtorchaudio and dynamic loader will handle it for us. 20 if path.exists(): ---> 21 torch.ops.load_library(path) 22 torch.classes.load_library(path) 23 # This import is for initializing the methods registered via PyBind11

/usr/local/lib/python3.7/dist-packages/torch/_ops.py in load_library(self, path) 108 # static (global) initialization code in order to register custom 109 # operators with the JIT. --> 110 ctypes.CDLL(path) 111 self.loaded_libraries.add(path) 112

/usr/lib/python3.7/ctypes/init.py in init(self, name, mode, handle, use_errno, use_last_error) 362 363 if handle is None: --> 364 self._handle = _dlopen(self._name, mode) 365 else: 366 self._handle = handle

OSError: libcudart.so.10.2: cannot open shared object file: No such file or directory

opened by janzuiderveld 0

Releases(v0.1.0-alpha)

v0.1.0-alpha(Oct 5, 2021)

pre-release v0.1.0-alpha
Source code(tar.gz)
Source code(zip)
Wav2CLIP.pt(46.69 MB)

Owner

Descript

GitHub Repository

Keras implementation of Normalizer-Free Networks and SGD - Adaptive Gradient Clipping

63 Sep 21, 2022

FS-Mol: A Few-Shot Learning Dataset of Molecules

FS-Mol is A Few-Shot Learning Dataset of Molecules, containing molecular compounds with measurements of activity against a variety of protein targets. The dataset is presented with a model evaluation

114 Dec 15, 2022

This package is for running the semantic SLAM algorithm using extracted planar surfaces from the received detection

Semantic SLAM This package can perform optimization of pose estimated from VO/VIO methods which tend to drift over time. It uses planar surfaces extra

125 Dec 02, 2022

Code for layerwise detection of linguistic anomaly paper (ACL 2021)

Layerwise Anomaly This repository contains the source code and data for our ACL 2021 paper: "How is BERT surprised? Layerwise detection of linguistic

6 Dec 07, 2022

An implementation of "MixHop: Higher-Order Graph Convolutional Architectures via Sparsified Neighborhood Mixing" (ICML 2019).

MixHop and N-GCN ⠀ A PyTorch implementation of "MixHop: Higher-Order Graph Convolutional Architectures via Sparsified Neighborhood Mixing" (ICML 2019)

393 Dec 13, 2022

Code for CVPR2021 "Visualizing Adapted Knowledge in Domain Transfer". Visualization for domain adaptation. #explainable-ai

Visualizing Adapted Knowledge in Domain Transfer @inproceedings{hou2021visualizing, title={Visualizing Adapted Knowledge in Domain Transfer}, auth

80 Dec 25, 2022

Computer Vision Script to recognize first person motion, developed as final project for the course "Machine Learning and Deep Learning"

Overview of The Code BaseColab/MLDL_FPAR.pdf: it contains the full explanation of our work Base Colab: it contains the base colab used to perform all

4 Jul 16, 2022

One-Shot Neural Ensemble Architecture Search by Diversity-Guided Search Space Shrinking

One-Shot Neural Ensemble Architecture Search by Diversity-Guided Search Space Shrinking This is an official implementation for NEAS presented in CVPR

19 Sep 08, 2022

UmlsBERT: Clinical Domain Knowledge Augmentation of Contextual Embeddings Using the Unified Medical Language System Metathesaurus

UmlsBERT: Clinical Domain Knowledge Augmentation of Contextual Embeddings Using the Unified Medical Language System Metathesaurus General info This is

71 Oct 25, 2022

Unofficial implementation of "Swin Transformer: Hierarchical Vision Transformer using Shifted Windows" (https://arxiv.org/abs/2103.14030)

Swin-Transformer-Tensorflow A direct translation of the official PyTorch implementation of "Swin Transformer: Hierarchical Vision Transformer using Sh

52 Dec 29, 2022

The official pytorch implemention of the CVPR paper "Temporal Modulation Network for Controllable Space-Time Video Super-Resolution".

This is the official PyTorch implementation of TMNet in the CVPR 2021 paper "Temporal Modulation Network for Controllable Space-Time VideoSuper-Resolu

95 Oct 24, 2022

A simple python library for fast image generation of people who do not exist.

Random Face A simple python library for fast image generation of people who do not exist. For more details, please refer to the [paper](https://arxiv.

170 Dec 15, 2022

Official PyTorch implementation of MX-Font (Multiple Heads are Better than One: Few-shot Font Generation with Multiple Localized Experts)

Introduction Pytorch implementation of Multiple Heads are Better than One: Few-shot Font Generation with Multiple Localized Expert. | paper Song Park1

97 Dec 23, 2022