List your company
ResearchFormerly WizWand

Music Modeling on J.S. Bach Chorales 16th notes (val)

0.22Validation NLLTonicNet
0.191360.384680.5780.77132Sep 12, 2018Nov 24, 2018Feb 5, 2019Apr 20, 2019Jul 2, 2019Sep 13, 2019Nov 26, 2019
Updated 8mo ago

Evaluation results

MethodLinks
TonicNetInput parts=CSATB, Repetition encoding (Z)=true, Data Augmentation=Tr+MM (Transposition + Major-to-Minor), Evaluation Protocol=NCL (No Chord Loss)2019.110.22
TonicNetInput parts=CSATB, Repetition encoding (Z)=true, Data Augmentation=Tr (Transposition), Evaluation Protocol=NCL (No Chord Loss)2019.110.224
TonicNetInput parts=CSATB, Repetition encoding (Z)=true, Data Augmentation=Tr+MM (Transposition + Major-to-Minor), Evaluation Protocol=Include chord loss2019.110.317
TonicNetInput parts=CSATB, Repetition encoding (Z)=true, Data Augmentation=Tr (Transposition), Evaluation Protocol=Include chord loss2019.110.321
Relative Transformer + cpsi + relative pitch and timeLayers=5L, Hidden size=512hs, Attention size=512att, Filter size=512ff, Relative distance (r)=256r, Heads=8h, cpsi=true, relative pitch and time=true2018.090.335
Music TransformerInput parts=SATB2019.110.335
Relative Transformer + concat positional sinusoids, instrument labels (cpsi)Layers=5L, Hidden size=512hs, Attention size=512att, Filter size=512ff, Relative distance (r)=256r, Heads=8h, cpsi=true2018.090.347
Relative TransformerLayers=5L, Hidden size=512hs, Attention size=512att, Filter size=512ff, Relative distance (r)=256r, Heads=8h2018.090.357
TF baseline + concat positional sinusoids, instrument labels (cpsi)Layers=5L, Hidden size=256hs, Attention size=256att, Filter size=1024ff, Heads=8h, cpsi=true2018.090.37
TransformerInput parts=CSATB, Evaluation Protocol=NCL (No Chord Loss)2019.110.394
TF baseline + concat positional sinusoids (cps)Layers=5L, Hidden size=256hs, Attention size=256att, Filter size=1024ff, Heads=8h, cps=true2018.090.398
Transformer (TF) baselineLayers=5L, Hidden size=256hs, Attention size=256att, Filter size=1024ff, Heads=8h2018.090.417
TonicNetInput parts=CSATB, Repetition encoding (Z)=true, Data Augmentation=None, Evaluation Protocol=Include chord loss2019.110.422
COCONET (CNN, chronological)Type=CNN, chronological, Layers=64L, Filter=128 3x3f2018.090.436
COCONETInput parts=SATB, Prediction Order=chronological2019.110.436
TonicNetInput parts=SATB, Repetition encoding (Z)=true2019.110.497
TransformerInput parts=CSATB, Evaluation Protocol=Include chord loss2019.110.503
TonicNetInput parts=CSB (Chords + Soprano + Bass)2019.110.516
TonicNetInput parts=S (Soprano)2019.110.521
TonicNetInput parts=SATB, Repetition encoding (Z)=false2019.110.523
TransformerInput parts=SATB2019.110.544
TonicNetInput parts=SB (Soprano + Bass)2019.110.555
TonicNetInput parts=CS (Chords + Soprano)2019.110.588
TonicNetInput parts=B (Bass)2019.110.716
TonicNetInput parts=C (Chords)2019.110.936