v5.2

• deeper audio feature network
• remove some filter projections

v5.1

config: deeper denoiser

v5.0

audio features: post conv proj

v4.0

• refactor: move data out of model (b5cbdae, 3d19f96)
• clean: documentation(41c7ec4, a4a75e0, 64573a1)
• feature: resample audio (2df1fc0)
• feature: remove timing (730ebc7)

v3.1

• refactor: generate_mapset to own function (1c48330)
• fix: integer timestamps for timing points (7ba48d7)
• feature: attempt to determine meter start when provided bpm (80e9c3b)
• feature: tag osu_dreamer on generated maps (4080fc0)

v3.0.3

• clean: quiet librosa load warning (f0fb5e9)

v3.0.2

• fix: more reliable audio loading (1888a2b)
• fix: zero length sliders (914671d)

v3.0.1

• fix: ensure hold starts precede hold ends (a54d660, edf28db)

v3.0

• feature: model conditioned on PLP or timing signal generated from timing points (b6af900, 2e3e5fa, 430a1d4)
• feature: beat snap when timing is provided (55f6f07)
• feature: proper bezier sliders + slider slides (77e2b0e, a6f84b4)
• refactor: osu_dreamer.model subpackage (b3b98c2)

v2.4

• refactor: extract data for training into separate module (8f4acda)