Will Dyson's Seinor Project Log

• Week 1:
  I decided to do my project on the subject of Neural Networks. Specificly, I want to work on detecting musical features with neural networks. Stuff like the beat of the music and what intstruments are playing at any given time. To learn some of the basics of NNs, I borowed and read Naturally Intelligent Systems by Caudill and Butler. See the References page for more info on any of the books I mention here.
• Week 2:
  Porter Schermerhorn and I decide to colaborate on this project. We will work together on finding a good Neural Network simulator and possibly adapting it for our needs. We don't yet know how we will be splitting the responsabilities. We think it would be cool if we could use the information our network extracts from music to do music visualisation of some kind.
• Week 3 - Week 10:
  Nothing
• Week 11:
  Lazyness about keeping this journal up to date. Lazyness on the project too, if truth be told. I've been doing research and playing with toy neural nets, but thats it. So a massive update ensues.

  I am using the NICO neural network toolkit http://www.speech.kth.se/NICO/ for my simulations. My goal is to engineer a network architechture and training regimen that can learn to classify a collection of musical instruments reliably.

  I am using samples of nine different instruments from The Electronic Music Studios at the University of Iowa. Each sample file contains an instrument playing a full octave. There are 110 sample files. Right now I have no plans to aquire more.

  I plan to pre-process the sound data by frequency-analysis (fourier transform) and logarithmic scaleing of the resulting power spectrum (since humans percieve the pitch to increase by a linear amount when the freqency doubles).

• Week 12:
  I have been working on getting a pre-processor built for my system. Porter has a version that works using the ALSA loopback device under linux. Not having that API on my platform of choice, I spent monday re-implementing it to use the Be MediaKit API for the reading of audio files.