markhanslip's picture
Update README.md
f59e669 verified
|
raw
history blame
834 Bytes
metadata
pretty_name: Mark Hanslip PhD Saxophone Data
tags:
  - tenor saxophone
  - improvisation
  - audio
license: cc-by-nc-nd-4.0
task_categories:
  - text-to-audio
  - audio-classification

A collection of recordings of private practice sessions on tenor saxophone, recorded in Leeds, UK in 2019 as part of my PhD project.

These recordings were initially created without a specific use-case in mind. They have since been used for the following tasks:

  • Audio Classification ("Timbral" and "Melodic Improv")
  • Generative Modelling with SampleRNN ("Tone Rows", "Scales and Arpeggios" and "Melodic Improv" worked best)
  • Timbre Transfer with RAVE (iil trained a RAVE model on the "Timbral" set)
  • Symbolic Modelling with modified Char-RNN (a composite of "Tone Rows", "Scales and Arpeggios" and "Melodic Improv" worked best)