BandNet: Neural Network-Based Multi-Instrument Music Composition
This research project introduces BandNet, a neural network-based system for multi-instrument Beatles-style MIDI music composition. By encoding musical scores using LSTM-RNN, the system addresses limitations of existing works and supports generating music scores for various purposes. Users can engage in the music assembly pipeline, selecting music clips and structuring songs based on a provided template. Results indicate subjective listening feedback on the music's similarity to The Beatles, perceived professionalism, and overall interest.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. Download presentation by click this link. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
E N D
Presentation Transcript
BandNet: A Neural Network-Based, Multi-Instrument Beatles-Style MIDI Music Composition Machine Yichao Zhou1,2 Wei Chu1Sam Young1,3 Xin Chen1 1Snap Inc. 2EECS, University of California, Berkeley. 3Herb Alpert School of Music, University of California, Los Angeles.
Computer Music Composition GOAL: Automatically generate music scores Generate musics for photo albums Build royalty-free sound libraries Harmonize vocal singing
The Beatles Dataset 123 MIDI files from The Beatles 4 labelled channels Melody (vocal part) Chords (harmony part) Bass Drum
Challenges Limitation of existing works (Magenta, FolkRNN, BachBot) Cannot support multiple instruments (Band!) Lack of song structure
Encoding the Score LSTM-RNN: How to deal with multiple instruments? RNN only takes a sequence as input NEW_NOTE E4 NEXT_CHANNEL NEW_NOTE C3 Melody NEXT_STEP CONTINUE_NOTE E4 NEXT_CHANNEL NEW_NOTE G3 NEW_NOTE C3 Bass NEXT_STEP
Music Assemble Pipeline 1. Users pick a seed for the each section; 2. LSTM-RNN generates music clips for each section; 3. Users listen to the clips and choose the one they like; 4. BandNet assembles the whole song according to a music structure template
Results (Subjective Listening) We asked subjects to listen to 14 testing songs and answer the questions (1-5): 1. Does it sound similar to the music from the Beatles? How likely is it that this music was professionally composed? How interesting is this music? 2. 3. Style Similarity Profession Interestingness
