Projects

Design/Implementation Projects

Implementation pointers

• Music Classification
  Objective: Classification of music melodies along three possible composers: Bach, Mozart and Chopin.
  Trick: Set up the global music duration as well as the shortest note duration.
  Input: a melody (Music21 or XML or MIDI format) of a fixed duration.
  Output: the most likely class.

• Counterpoint Accompaniment
  Objective: Generation of the counterpoint accompaniment of a melody.
  Trick: Iterative generation by a feedforward architecture generating time slice by time slice. Compare with the standard (course) architecture and the triple model (course) architecture.
  Input: a melody (Music21 or XML or MIDI format) of a fixed duration.
  Output: the three accompaniment voices.

• Chord Accompaniment
  Objective: Generation of the chords harmony accompaniment of a melody.
  Trick: Representation of chords.
  Input: a melody (Music21 or XML or MIDI format) of a fixed duration.
  Output: the accompanying sequence of chords.

• Personal Proposal

Article Presentation Projects

• Latent Constraints Learning to Generate Conditionally from Unconditional Generative Models
• MUSIC TRANSFORMER - GENERATING MUSIC WITH LONG-TERM STRUCTURE
• SONG FROM PI: A MUSICALLY PLAUSIBLE NETWORK FOR POP MUSIC GENERATION
• JazzGAN - Improvising with Generative Adversarial Networks
• An End to End Model for Automatic Music Generation: Combining Deep Raw and Symbolic Audio Networks
• What do these 5,599,881 parameters mean? An analysis of a specific LSTM music transcription model, starting with the 70,281 parameters of its softmax layer
• A Universal Music Translation Network
• A Recurrent Neural Network for Musical Structure Processing and Expectation
• A Hierarchical Latent Vector Model for Learning Long-Term Structure in Music

Allocation/Choice

• Estevan Barbará, Article: What do these 5,599,881 parameters mean? An analysis of a specific LSTM music transcription model, starting with the 70,281 parameters of its softmax layer

Presentation Agenda

• 16 December 2019
  • Estevan Barbará, Article: What do these 5,599,881 parameters mean? An analysis of a specific LSTM music transcription model, starting with the 70,281 parameters of its softmax layer Presentation