Ehaschia / osu-mapper

An automatic osu! map generator using deep learning method

osu-mapper

osu!mapper

TODO

osu Format File [Liwen]

• Slider end time
• Slider points((X, Y) in the path)
• v6 ~ v14(need to generalize, output format: v14)
• HitSounds
• Lines under editor

Model [Jun]

Use RNN-like model. State h_0 = some hyper-parameters in osu file and features of music. Input X = features of music segmentations. Output = state h_1 .. h_n, which can be mapped to osu file(use a matrix and a tool).

• Space modeling
  • Map state to a 2d vector
• Model of slider(length should be legal)  - Length don't need to be legal(velocity is not fixed)
• Model with more prior
  • e.g. points are related

Music [Jiong]

• Basic concepts
  • timbre 音色: 音色主要决定于声音频谱对人的刺激，但也决定于波形、声压、频谱的频率位置和频谱对人的时间性刺激
  • rhythem 节奏:
  • pitch information 音高:
• Process(how?)
  • Feature: MFCC (Tutorial)
    • A survey
  • Relative Package:
    • python speech features
    • librosa - pydub
• Music model(for machine learning)