YOLOX_AUDIO is an audio event detection model based on YOLOX

Last update: Dec 19, 2022

Related tags

Deep Learning YOLOX_AUDIO

Overview

Introduction

YOLOX_AUDIO is an audio event detection model based on YOLOX, an anchor-free version of YOLO. This repo is an implementated by PyTorch. Main goal of YOLOX_AUDIO is to detect and classify pre-defined audio events in multi-spectrogram domain using image object detection frameworks.

Updates!!

【2021/11/15】 We released YOLOX_AUDIO to public

Quick Start

Installation

Step1. Install YOLOX_AUDIO.

git clone https://github.com/intflow/YOLOX_AUDIO.git
cd YOLOX_AUDIO
pip3 install -U pip && pip3 install -r requirements.txt
pip3 install -v -e .  # or  python3 setup.py develop

Step2. Install pycocotools.

pip3 install cython; pip3 install 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'

Data Preparation

Step1. Prepare audio wavform files for training. AUDIO_DATAPATH/wav

Step2. Write audio annotation files for training. AUDIO_DATAPATH/label.json

{
    "00000.wav": {
        "speaker": [
            "W",
            "M",
            "C",
            "W"
        ],
        "on_offset": [
            [
                1.34425,
                2.4083125
            ],
            [
                4.0082708333333334,
                4.5560625
            ],
            [
                6.2560416666666665,
                7.956104166666666
            ],
            [
                9.756083333333333,
                10.876624999999999
            ]
        ]
    },
    "00001.wav": {
        "speaker": [
            "W",
            "M",
            "C",
            "M",
            "W",
            "C"
        ],
        "on_offset": [
            [
                1.4325416666666666,
                2.7918958333333332
            ],
            [
                2.1762916666666667,
                4.109729166666667
            ],
            [
                7.109708333333334,
                8.530916666666666
            ],
            [
                8.514125,
                9.306104166666668
            ],
            [
                12.606083333333334,
                14.3345625
            ],
            [
                14.148958333333333,
                15.362958333333333
            ]
        ]
    },
    ...
}

Step3. Convert audio files into spectrogram images.

python tools/json_gen_audio2coco.py

Please change the dataset path and file names for your needs

root = '/data/AIGC_3rd_2021/GIST_tr2_veryhard5000_all_tr2'
os.system('rm -rf '+root+'/img/')
os.system('mkdir '+root+'/img/')
wav_folder_path = os.path.join(root, 'wav')
img_folder_path = os.path.join(root, 'img')
train_label_path = os.path.join(root, 'tr2_devel_5000.json')
train_label_merge_out = os.path.join(root, 'label_coco_bbox.json')

Training

Step1. Change Data loading path of exps/yolox_audio__tr2/yolox_x.py

        self.train_path = '/data/AIGC_3rd_2021/GIST_tr2_veryhard5000_all_tr2'
        self.val_path = '/data/AIGC_3rd_2021/tr2_set_01_tune'
        self.train_ann = "label_coco_bbox.json"
        self.val_ann = "label_coco_bbox.json"

Step2. Begin training:

python3 tools/train.py -expn yolox_audio__tr2 -n yolox_audio_x \
-f exps/yolox_audio__tr2/yolox_x.py -d 4 -b 32 --fp16 \
-c /data/pretrained/yolox_x.pth

-d: number of gpu devices
-b: total batch size, the recommended number for -b is num-gpu * 8
-f: path of experiement file
--fp16: mixed precision training
--cache: caching imgs into RAM to accelarate training, which need large system RAM.

We are encouraged to use pretrained YOLOX model for the training. https://github.com/Megvii-BaseDetection/YOLOX

Inference

Run following demo_audio.py

python3 tools/demo.py --demo image -expn yolox_audio__tr2 -n yolox_audio_x \
-f exps/yolox_audio__tr2/yolox_x.py \
-c YOLOX_outputs/yolox_audio__tr2/best_ckpt.pth \
--path /data/AIGC_3rd_2021/GIST_tr2_100/img/ \
--save_folder /data/yolox_out \
--conf 0.2 --nms 0.65 --tsize 256 --save_result --device gpu

From the demo_audio.py you can get on-offset VAD time and class of each audio chunk.

References

YOLOX baseline implemented by PyTorch: YOLOX

 @article{yolox2021,
  title={YOLOX: Exceeding YOLO Series in 2021},
  author={Ge, Zheng and Liu, Songtao and Wang, Feng and Li, Zeming and Sun, Jian},
  journal={arXiv preprint arXiv:2107.08430},
  year={2021}
}

Librosa for audio feature extraction: librosa

McFee, Brian, Colin Raffel, Dawen Liang, Daniel PW Ellis, Matt McVicar, Eric Battenberg, and Oriol Nieto. “librosa: Audio and music signal analysis in python.” In Proceedings of the 14th python in science conference, pp. 18-25. 2015.

Acknowledgement

This work was supported by the Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No. 2021-0-00014).

YOLOX_AUDIO is an audio event detection model based on YOLOX

Related tags

Overview

Introduction

Updates!!

Quick Start

References

Acknowledgement

Owner

intflow Inc.

This repository contains the code, data, and models of the paper titled "XL-Sum: Large-Scale Multilingual Abstractive Summarization for 44 Languages" published in Findings of the Association for Computational Linguistics: ACL 2021.

Alignment Attention Fusion framework for Few-Shot Object Detection

Hough Transform and Hough Line Transform Using OpenCV

Text-to-Music Retrieval using Pre-defined/Data-driven Emotion Embeddings

A Number Recognition algorithm

Pseudo lidar - (CVPR 2019) Pseudo-LiDAR from Visual Depth Estimation: Bridging the Gap in 3D Object Detection for Autonomous Driving

Boundary-aware Transformers for Skin Lesion Segmentation

Aircraft design optimization made fast through modern automatic differentiation

Deep Two-View Structure-from-Motion Revisited

A annotation of yolov5-5.0

This is a custom made virus code in python, using tkinter module.

Modifications of the official PyTorch implementation of StyleGAN3. Let's easily generate images and videos with StyleGAN2/2-ADA/3!

Sentiment analysis translations of the Bhagavad Gita

Sequence to Sequence (seq2seq) Recurrent Neural Network (RNN) for Time Series Forecasting

Implementation of the ALPHAMEPOL algorithm, presented in Unsupervised Reinforcement Learning in Multiple Environments.

Official Matlab Implementation for "Tiny Obstacle Discovery by Occlusion-aware Multilayer Regression", TIP 2020

Code release for "Detecting Twenty-thousand Classes using Image-level Supervision".

A implemetation of the LRCN in mxnet

Python periodic table module

Learning Tracking Representations via Dual-Branch Fully Transformer Networks