Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Musical Deep Learning

No description
by

Ben Smith

on 19 June 2017

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Musical Deep Learning

Indiana University-Purdue University-Indianapolis
School of Engineering & Technology
Dpt. of Music and Arts Technology, Dpt. of Computer Graphics Technology
Musical Deep Learning
Benjamin D. Smith
MUME 2017, Atlanta, GA
June 19-20
Musical possibilities of the Conditional Restricted Boltzmann Machine (CRBM)
Zip (M = 387.03,σ = 64.42)
LZW (M = 509.53,σ = 31.82)
Shannon's Entropy
Shannon's Entropy vs. LZW
Complexity of Bach
J.S. Bach Suites for Solo Violoncello, BWV 1007-1012
6 Suites, 36 movements
30222 note events
Training corpus
Primarily monophonic, implied harmonic structures
Complexity of CRBM generation
Results
http://www.uoguelph.ca/~gwtaylor/thesis/
CRBM, Graham Taylor
Feature Sets
MIDI pitch–single feature
Derivatives of pitch (aka velocity, accel, jerk, etc.)
Piano roll (aka 'one hot' binary vector)
Pitch class ('one hot')
Interval tokens ('one hot')
FFT–windowed note sequences, hop=1 note
Best musical encoding to leverage CRBM?
Midi Pitch alone
Pitch + 12 derivatives
Piano Roll
Piano Roll + 12 Pitch Classes
Interval tokens
FFT (window size=8, hop=1)
Higher order (n=7), Piano Roll
CRBM with all features
Shannon's Entropy vs. LZW
All features, order=7
All features, order=9
Higher orders vs. Bach
(2009 thesis, student of Hinton)
Extensively researched in musicological literature
Lempel-Ziv string compression
1000 note sequences
(tried interpolation, had no discernible effect)
All examples: Nv=Nh=100, order = 3
Full transcript