add maestro dataset PR #13

CONTRIBUTING.md

@@ -9,11 +9,11 @@ You can look at the issues and help to solve them or you can add some missing ar
 Here are the steps to follow for adding one (or multiple) article:
 1. Check that the article is not already in the [dl4m.bib](dl4m.bib) file.
 2. Fork the repo.
-3. Add the desired bib entry at the beginning of [dl4m.bib](dl4m.bib). Take care to fill all this field for each bib entry:
-    - Bib entry type (inproceedings, article, techreport, unpublished,...)
+3. Add the desired bib entry at the beginning of [dl4m.bib](dl4m.bib). Take care to fill all this field for each bib entry (if there is no information about a field please indicate it as `fieldName = {No}`):
+    - Bib entry type (inproceedings, article, techreport, unpublished,...) (N.B.: Indicate arxiv article as @unpublished and if you know of a possible submission use `note = {Submitted to "NameOfJournalOrConference"}`)
     - Bib key (in the form AuthorlastnameYear, e.g. `Snow1999`
-    - title
-    - author
+    - title (N.B.: Make sure the title is not written in all caps and that each word does not start with a capital letter.)
+    - author (N.B.: Check that all authors in the pdf files are present in the bib author field and in the same order. Indeed automatic bib tools tend to mess up the order or to forget some authors.)
     - year
     - booktitle or journal
     - dataset (e.g. `dataset = {Inhouse & [Jamendo](http://www.mathieuramona.com/wp/data/jamendo/) & [RWC](https://staff.aist.go.jp/m.goto/RWC-MDB/)},`)

README.md

@@ -195,6 +195,7 @@ However, these surveys do not cover music information retrieval tasks that are i
 | 2018 | [MuseGAN: Multi-track sequential generative adversarial networks for symbolic music generation and accompaniment](https://arxiv.org/pdf/1709.06298.pdf) | [GitHub](https://github.com/salu133445/musegan) |
 | 2018 | [Music transformer: Generating music with long-term structure](https://arxiv.org/pdf/1809.04281.pdf) | No |
 | 2018 | [Music theory inspired policy gradient method for piano music transcription](https://nips2018creativity.github.io/doc/music_theory_inspired_policy_gradient.pdf) | No |
+| 2019 | [Enabling factorized piano music modeling and generation with the MAESTRO dataset](https://arxiv.org/abs/1810.12247) | No |
 | 2019 | [Generating Long Sequences with Sparse Transformers](https://arxiv.org/pdf/1904.10509.pdf) | [GitHub](https://github.com/openai/sparse_attention) |
 address = {Montreal, Canada} | [Music theory inspired policy gradient method for piano music transcription](https://nips2018creativity.github.io/doc/music_theory_inspired_policy_gradient.pdf) | No |
 reproducible = {No} | [Generating Long Sequences with Sparse Transformers](https://arxiv.org/pdf/1904.10509.pdf) | [GitHub](https://github.com/openai/sparse_attention) |
@@ -250,15 +251,15 @@ Each entry in [dl4m.bib](dl4m.bib) also displays additional information:
 
 ## Statistics and visualisations
 
-- 164 papers referenced. See the details in [dl4m.bib](dl4m.bib).
+- 165 papers referenced. See the details in [dl4m.bib](dl4m.bib).
 There are more papers from 2017 than any other years combined.
 Number of articles per year:
 ![Number of articles per year](fig/articles_per_year.png)
-- If you are applying DL to music, there are [352 other researchers](authors.md) in your field.
+- If you are applying DL to music, there are [356 other researchers](authors.md) in your field.
 - 34 tasks investigated. See the list of [tasks](tasks.md).
 Tasks pie chart:
 ![Tasks pie chart](fig/pie_chart_task.png)
-- 52 datasets used. See the list of [datasets](datasets.md).
+- 53 datasets used. See the list of [datasets](datasets.md).
 Datasets pie chart:
 ![Datasets pie chart](fig/pie_chart_dataset.png)
 - 30 architectures used. See the list of [architectures](architectures.md).
@@ -321,6 +322,7 @@ The list of conferences, journals and aggregators used to gather the proposed ma
 - [Jordi Pons](http://www.jordipons.me/) ([GitHub](https://github.com/jordipons))
 - Mirza Zulfan ([GitHub](https://github.com/mirzazulfan)) for the logo
 - [Devin Walters](https://github.com/devn)
+- https://github.com/LegendJ
 
 [Go back to top](https://github.com/ybayle/awesome-deep-learning-music#deep-learning-for-music-dl4m-)
 

authors.md

@@ -62,6 +62,8 @@
 - Eck, Douglas
 - Ehmann, Andreas F.
 - Ellis, Daniel P. W.
+- Elsen, Erich
+- Engel, Jesse
 - Evangelopoulos, Georgios
 - Ewert, Sebastian
 - Fan, Zhe-Cheng
@@ -246,6 +248,7 @@
 - Richard, Gaël
 - Riedmiller, Martin
 - Rigaud, François
+- Roberts, Adam
 - Robinson, John
 - Roma, Gerard
 - Rosasco, Lorenzo
@@ -281,6 +284,7 @@
 - Southall, Carl
 - Stables, Ryan
 - Stasiak, Bartłomiej
+- Stasyuk, Andriy
 - Stoller, Daniel
 - Sturm, Bob L.
 - Su, Hong

datasets.md

@@ -34,6 +34,7 @@ Please refer to the list of useful acronyms used in deep learning and music: [ac
 - [Lakh Pianoroll Datase](https://github.com/salu133445/musegan/blob/master/docs/dataset.md)
 - [Last.fm](https://www.last.fm/)
 - [LyricFind](http://lyricfind.com/)
+- [MAESTRO](https://magenta.tensorflow.org/datasets/maestro/)
 - [MAPS](http://www.tsi.telecom-paristech.fr/aao/en/2010/07/08/maps-database-a-piano-database-for-multipitch-estimation-and-automatic-transcription-of-music/)
 - [MIR-1K](https://sites.google.com/site/unvoicedsoundseparation/mir-1k)
 - [MSD](https://labrosa.ee.columbia.edu/millionsong/)

dl4m.bib

@@ -2159,6 +2159,39 @@ @inproceedings{Li2018
   year = {2018}
 }
 
+@inproceedings{Hawthorne2019,
+  activation = {No},
+  address = {New Orleans, USA},
+  architecture = {No},
+  author = {Hawthorne, Curtis and Stasyuk, Andriy and Roberts, Adam and Simon, Ian and Huang, Cheng-Zhi Anna and Dieleman, Sander and Elsen, Erich and Engel, Jesse and Eck, Douglas},
+  batch = {No},
+  booktitle = {ICLR},
+  code = {No},
+  computationtime = {No},
+  dataaugmentation = {pitch-shift {+-0.1 semitones} & compression {0 - 100} & EQ {32 - 4096} & Reverb & {0 - 70} & Pink-noise {0 - 0.04}},
+  dataset = {[MAESTRO](https://magenta.tensorflow.org/datasets/maestro/)},
+  dimension = {1D},
+  dropout = {No},
+  epochs = {No},
+  framework = {No},
+  gpu = {No},
+  input = {MIDI},
+  layers = {6},
+  learningrate = {No},
+  link = {https://arxiv.org/abs/1810.12247},
+  loss = {No},
+  metric = {MSE & Precision & Recall & F1},
+  momentum = {No},
+  month = {May},
+  note = {Wave2Midi2Wave},
+  optimizer = {No},
+  pages = {1--12},
+  reproducible = {No},
+  task = {Transcription},
+  title = {Enabling factorized piano music modeling and generation with the MAESTRO dataset},
+  year = {2019}
+}
+
 @unpublished{Child2019,
   activation = {No},
   architecture = {Transformer},

dl4m.tsv

@@ -162,4 +162,5 @@ Year	Entrytype	Title	Author	Link	Code	Task	Reproducible	Dataset	Framework	Archit
 2018	inproceedings	MuseGAN: Multi-track sequential generative adversarial networks for symbolic music generation and accompaniment	Dong, Hao-Wen and Hsiao, Wen-Yi and Yang, Li-Chia and Yang, Yi-Hsuan	https://arxiv.org/pdf/1709.06298.pdf	https://github.com/salu133445/musegan	Composition	No	[Lakh Pianoroll Datase](https://github.com/salu133445/musegan/blob/master/docs/dataset.md)	No	GAN & CNN	No	No	No	No	Piano-roll	1D	ReLU & Leaky ReLU	No	No	Adam	1 Tesla K40m	
 2018	unpublished	Music transformer: Generating music with long-term structure	Huang, Cheng-Zhi Anna and Vaswani, Ashish and Uszkoreit, Jakob and Shazeer, Noam and Simon, Ian and Hawthorne, Curtis and Dai, Andrew M. and Hoffman, Matthew D. and Dinculescu, Monica and Eck, Douglas	https://arxiv.org/pdf/1809.04281.pdf		Polyphonic music sequence modelling	No	[J.S. Bach chorales dataset](https://github.com/czhuang/JSB-Chorales-dataset) & [Piano-e-Competition dataset (competition history)](http://www.piano-e-competition.com/)	No	Transformer & RNN & tensor2tensor	0.1	1	No	Time Stretches & pitch transcription	MIDI	1D	No	No	0.1	No	No	
 2018	inproceedings	Music theory inspired policy gradient method for piano music transcription	Li, Juncheng and Qu, Shuhui and Wang, Yun and Li, Xinjian and Das, Samarjit and Metze, Florian	https://nips2018creativity.github.io/doc/music_theory_inspired_policy_gradient.pdf	No	Transcription	No	[MAPS](http://www.tsi.telecom-paristech.fr/aao/en/2010/07/08/maps-database-a-piano-database-for-multipitch-estimation-and-automatic-transcription-of-music/)		CNN & RNN	No	8	No	No	Log Mel-spectrogram with 48 bins per octave and 512 hop-size and 2018 window size and 16 kHz sample rate	2D	No	binary cross-entropy	0.0006	Adam	No	
+2019	inproceedings	Enabling factorized piano music modeling and generation with the MAESTRO dataset	Hawthorne, Curtis and Stasyuk, Andriy and Roberts, Adam and Simon, Ian and Huang, Cheng-Zhi Anna and Dieleman, Sander and Elsen, Erich and Engel, Jesse and Eck, Douglas	https://arxiv.org/abs/1810.12247	No	Transcription	No	[MAESTRO](https://magenta.tensorflow.org/datasets/maestro/)	No	No	No	No	No	pitch-shift {+-0.1 semitones} & compression {0 - 100} & EQ {32 - 4096} & Reverb & {0 - 70} & Pink-noise {0 - 0.04}	MIDI	1D	No	No	No	No	No	
 2019	unpublished	Generating Long Sequences with Sparse Transformers	Rewon Child and Scott Gray and Alec Radford and Ilya Sutskever	https://arxiv.org/pdf/1904.10509.pdf	https://github.com/openai/sparse_attention	Audio generation	No	[413 hours of recorded solo piano music](http://papers.nips.cc/paper/8023-the-challenge-of-realistic-music-generation-modelling-raw-audio-at-scale-supplemental.zip)	Tensorflow	Transformer	0.25	No	120	No	Raw Audio	1D	No	No	0.00035	Adam	8 NVIDIA Tesla V100