Contact
CoCalc Logo Icon
StoreFeaturesDocsShareSupport News AboutSign UpSign In
| Download

You know...

Views: 68
1
\documentclass[12pt]{article}
2
\usepackage{fullpage}
3
4
\title{CSE 546 Autumn 2018 - Proposal \\ Project title: \emph{Larghissimo}}
5
\author{Charles Godfrey \\ Graham Gordon}
6
7
8
\begin{document}
9
\maketitle
10
11
We propose to use the FreeMIDI \cite{freemidi} dataset of MIDI files together with neural networks to train a computer to predict melodies and potentially harmonies. As an application, we intend to use our algorithm to generate its own melodies (and harmonies) given a sample melody as input.
12
13
We propose to learn about several neural network techniques, including LSTMs, RNNs, and CNNs, which are applied in the references. We have not studied these algorithms before. We will seek out the necessary references to learn about them when the time comes.
14
15
We propose to study the results of our algorithm with an online user study. Here is one idea for a study we are considering, along the lines of BachBot \cite{bachbot}: we would post a series of audio questions online. For each question, the user would listen to melody audio from two MIDI files, one composed by \emph{Larghissimo}, the other composed by a human being. The user would then choose which audio sample they believe was composed by a human.
16
17
By the project milestone, we plan to have
18
\begin{itemize}
19
\item studied \cite{deepbach, spectrogram, sampleRNN, jambot},
20
\item processed our data into a usable format, and
21
\item implemented at least one algorithm, perhaps a simplified form of a method used in the references.
22
\end{itemize}
23
24
25
26
\bibliographystyle{abbrv}
27
\bibliography{../the}
28
29
30
31
\end{document}
32
33