Datasets - HackMD

# Datasets for AIM ## [MusicNet](https://zenodo.org/record/5120004#.Yhxr0-jMJBA) - **File Format**: (WAV, CSV, MIDI) - **Data Included**: (330 freely-licensed classical music recordings, together with over 1 million annotated labels indicating the precise time of each note in every recording,) - **Dataset Size**: (32.1 TB) ## [Cocochorales](https://magenta.tensorflow.org/datasets/cocochorales) - **File Format**: (e.g., WAV, MP3, MIDI) - **Data Included**: (MIDI and MIDI generated audio paris of quartet instruments) - **Dataset Size**: (4Tb) - ## [University of Rochester Multi-Modal Music Performance (URMP) Dataset](https://datadryad.org/stash/dataset/doi:10.5061/dryad.ng3r749) - **File Format**: (WAV, MP4, MIDI) - **Data Included**: ( 44 simple multi-instrument classical music pieces assembled from coordinated but separately recorded performances of individual tracks) - **Dataset Size**: (12.5 GB) ## [Pompeu Fabra University Multimodal String Quartet Performance Dataset](https://www.upf.edu/web/mtg/quartet-dataset) - **File Format**: Specialized/Proprietary format - **Data Included**: String Quarter performance (Audio, Score-Performance Alignment, Wired/Wireless Motion Capture) - **Dataset Size**: Unknown ## [3D-video Point Cloud Musicians Dataset](https://zenodo.org/doi/10.5281/zenodo.4812951) - **File Format**: (PLY / Polygon File) - **Data Included**: 12 1-hour recordings of 3D video of musicians playing each of **cello, doublebass, guitar, saxophone, and violin**. - **Dataset Size**: (14.9 GB) - ## [AudioSet-2m](http://research.google.com/audioset/) - **File Format**: (Tfrecords) - **Data Included**: (2.1 million annotated videos, 5.8 thousand hours of audio, 527 classes of annotated sounds) - **Dataset Size**: ~14TB ## [MAESTRO](https://magenta.tensorflow.org/datasets/maestro) - **File Format**: ( WAV, MIDI) - **Data Included**: ( 200 hours of paired audio and MIDI recordings of Piano playing) - **Dataset Size**: (...) ## [GuitarSet](https://guitarset.weebly.com) - **File Format**: (WAV, .jams) - **Data Included**: (360 Guitar playing excerpts that are close to 30 seconds in length) - **Dataset Size**: (8GB) ## [FMA: A Dataset for Music Analysis](https://github.com/mdeff/fma) - **File Format**: (mp3) - **Data Included**: (fma_small.zip: 8,000 tracks of 30s, 8 balanced genres (GTZAN-like) (7.2 GiB) fma_medium.zip: 25,000 tracks of 30s, 16 unbalanced genres (22 GiB) fma_large.zip: 106,574 tracks of 30s, 161 unbalanced genres (93 GiB) fma_full.zip: 106,574 untrimmed tracks, 161 unbalanced genres (879 GiB)) - **Dataset Size**: (varying) ## [Solos](https://juanmontesinos.com/Solos/) - **File Format**: (MKV, MP4, WEBM) - **Data Included**: ~750 total recordings, true total can vary due to number of videos. The same instruments are in this dataset as the URMP Dataset (violin, viola, cello, double bass, flute, oboe, clarinet, bassoon, saxophone, trumpet, horn, trombone, tuba) - These are recordings uploaded to YouTube that have decent audio and camera quality. Instrument sound is isolated and the videos show musicians in good angles to determine posture. - **Dataset Size**: As a preview, downloaded only Cello, Violin, and Viola datasets, which came out to be ~15.1 GB ## [Trios](https://zenodo.org/record/6797837) - **File Format**: (MID, WAV, PDF) - **Data Included**: Recordings of 5 songs, with each individual instrument having a distinct recorded audio, synthesized audio, and a midi file. Instruments included are clarinet, viola, piano, violin, cello, French horn, trumpet, bassoon, alto sax, and drums) - There are also pdfs of the score being played - **Dataset Size**: There are only 5 total songs; 129 MB ## [DexYCB](https://dex-ycb.github.io/) - **File Format**: Video, Metadata - **Data Included**: Video of subjects' hands interacting with objects with relevant 3D data of position and orientation included. - **Dataset Size**: 119GB compressed (10 subjects with 12GB of data each + ) ## [kunstderfuge](https://www.kunstderfuge.com) - **File Format**: MIDI - **Data Included**: 20,000+ (1,000+ composers) - **Dataset Size**: 2.05GB ## [NSynth](https://magenta.tensorflow.org/datasets/nsynth) - **File Format**: (WAV), metadata - **Data Included**: 300,000 .wav files (split into training, validation, and test sets) containing single articulations of an instrument (either acoustic or electric). For each .wav file, 1 corresponding .json and .tfrecord file are also included, containing metadata and features about the note, respectively. - **Dataset Size**: The test set when downloaded was 400MB compressed for 4096 examples. Assume the entire dataset is on the order of 30-40GB. ## [AVSpeech](https://looking-to-listen.github.io/avspeech/) ## [good Sounds](https://zenodo.org/records/4588740#.YFDoDdyCFPY) - for tonal models # Others ## [Million Song Dataset (MSD)](https://labrosa.ee.columbia.edu/millionsong/) - **File Format**: (HDF5) - **Data Included**: (Metadata and audio features for a million contemporary tracks) - **Dataset Size**: (Approximately 280 GB) ## [Lakh MIDI Dataset](https://colinraffel.com/projects/lmd/) - **File Format**: (MIDI) - **Data Included**: (MIDI files matched to the Million Song Dataset tracks) - **Dataset Size**: (Approximately 45 GB) ## [FMA: A Dataset for Music Analysis](https://github.com/mdeff/fma) - **File Format**: (MP3) - **Data Included**: (Tracks annotated with genres, metadata, and tags) - **Dataset Size**: (Varies, up to 879 GB for full dataset) ## [GTZAN Genre Collection](https://www.kaggle.com/datasets/andradaolteanu/gtzan-dataset-music-genre-classification) - **File Format**: (WAV) - **Data Included**: - **Dataset Size**: (Approximately 1.41 GB) ## [The NSynth Dataset](https://magenta.tensorflow.org/datasets/nsynth) - **File Format**: (WAV, JSON) - **Data Included**: (300,000 musical notes with annotated pitch, velocity, instrument) - **Dataset Size**: (Approximately 76 GB) ## [IRMAS](https://www.upf.edu/web/mtg/irmas) - **File Format**: (WAV) - **Data Included**: (Instrument recognition in polyphonic music) - **Dataset Size**: (Approximately 11 GB) ## [Groove MIDI Dataset (Magenta)](https://magenta.tensorflow.org/datasets/groove) - **File Format**: (MIDI) - **Data Included**: (Drum performance data including MIDI files and audio) - **Dataset Size**: (Approximately 2.8 GB) ## [GiantSteps Tempo and Key Dataset](http://giantsteps-data.eecs.qmul.ac.uk/) - **File Format**: (Audio, Annotations) - **Data Included**: (Data for tempo detection and key estimation) - **Dataset Size**: (Approximately 2 GB) ## [RWC Music Database](https://staff.aist.go.jp/m.goto/RWC-MDB/) - **File Format**: (WAV, MIDI) - **Data Included**: (Music for perception and performance analysis) - **Dataset Size**: (Approximately 200 GB) ## [MedleyDB](http://medleydb.weebly.com/) - **File Format**: (Multitrack Audio, Annotations) - **Data Included**: (Annotated multitrack audio for music research) - **Dataset Size**: (Approximately 247.8 TB) ## [MIR-1K](https://zenodo.org/records/3532216) - **File Format**: (WAV) - **Data Included**: (dataset for singing voice separation) - **Dataset Size**: (Approximately 2.2 GB) ## [Jazz Solo Dataset](http://jazzomat.hfm-weimar.de/dbformat/dboverview.html) - **File Format**: (MIDI, CSV) - **Data Included**: (Transcriptions of jazz solos) - **Dataset Size**: (Approximately 2 GB) ## [Vienna 4x22 Piano Corpus](https://www.kaggle.com/datasets/ashkhagan/the-vienna-4x22-piano-corpus) - **File Format**: (WAV) - **Data Included**: (Recordings of piano performances) - **Dataset Size**: (Approximately 2.1MB)