Project Report

Introduction

Motivation
Previous works

Problem Definition

Input format
Output format

Model Designs

Features

MFCC(+delta)
cqt(+delta)
raw input

Convnet

Based on features mentioned above

Baseline

MFCC + delta1 + delta2 -> Fit Gaussian -> Parameters -> features -> One vs All logistic regression

Evaluation metrics

Accuracy
F-micro
F-macro

Experiments

Dataset

Format
Summary

Label Generation

From activations to label

Training procedures

Dataset split

Preprocessing

Channel-wise global normalization

Convnet Architecture

Result

Discussion

Conlusion