GithubHelp home page GithubHelp logo

satroan / badc-2017 Goto Github PK

View Code? Open in Web Editor NEW

This project forked from karolpiczak/badc-2017

0.0 1.0 0.0 11.31 MB

Bird Audio Detection challenge submission using an ensemble of convolutional neural networks

Python 100.00%

badc-2017's Introduction

BADC-2017

Bird Audio Detection challenge submission using an ensemble of convolutional neural networks


Challenge info: http://machine-listening.eecs.qmul.ac.uk/bird-audio-detection-challenge/

The task is to design a system that, given a short audio recording, returns a binary decision for the presence/absence of bird sound (bird sound of any kind).


Pretty straightforward implementation of convolutional detectors for this task, nothing too fancy. Data augmentation done with:

  • noise corruption of training files,
  • averaging of two training spectrograms at random,
  • time-shifts of whole recordings,
  • permutations of spectrogram segments.

Final submission consists of averaged predictions from 6 nets:

run-A

High resolution spectrogram convnet:

  • trained on whole dataset,
  • 5 conv/pooling LeakyReLU layers with batch normalization and delta channel,
  • global max pooling before output.

A

run-B

Same as run-A, different random seed.

B

run-A_ds

Same as run-A, but low resolution (downsampled).

A_ds

run-B_ds

Same as run-A_ds, different random seed.

B_ds

run-23

Low resolution spectrogram convnet.

When compared to runs A_ds/B_ds:

  • less filters,
  • no L2 regularization,
  • more epochs,
  • faster learning rate decay,
  • less dropout.

Originally used shuffle_spec() with 5 segments (spec length of 475) and less noise augmentations to generate the predictions.

23

run-26

High resolution spectrogram convnet.

When compared to runs A/B:

  • more filters,
  • lower regularization,
  • lower learning rate,
  • more epochs.

26

badc-2017's People

Contributors

karolpiczak avatar

Watchers

James Cloos avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    ๐Ÿ–– Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. ๐Ÿ“Š๐Ÿ“ˆ๐ŸŽ‰

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google โค๏ธ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.