Results for ""
Music-generation datasets 4 Min Read Sep 22, 2022
Leading music-generation datasets in 2022
Technology has been used to create music since the 1700s, from Mozart's use of music dice tables (Kingberger 1757) to Ada Lovelace, the first computer programmer, who foresaw computers being able to create music.
Published By : Dr Nivash Jeevanandam
Virtual assistants, in-car navigation, and creating new melodies based on music datasets for machine learning are some examples of machine learning applications. It uses audio data and is frequently applied in a variety of industries.
AI has long been used to create synthetic music. Nonetheless, a watershed moment occurred when researchers introduced AI to music informatics. Artificial music intelligence researchers are currently capitalizing on recent advances in AI, ML, and analytics to create models that can compose music at the same level as humans.
A model capable of generating music should be able to recognize patterns in a large number of music datasets and create original music inspired by the datasets. Datasets of music for A model capable of generating music should be able to recognize patterns in a large number of music datasets and create original music inspired by the datasets.
The data preparation process will ensure that the datasets are compatible with the machine learning algorithm and that the best audio collection mechanism is chosen.
An AI or ML model is only as good as the information it uses. This article examines the 2022 music industry's most popular music generation datasets.
Gillick et al. introduce Groove (Groove MIDI Dataset) in Learning to Groove with Inverse Sequence Transformations.
The Groove MIDI Dataset (GMD) contains 13.6 hours of aligned MIDI and (synthesized) audio of tempo-aligned expressive drumming performed by humans. The dataset consists of 1,150 MIDI files and over 22,000 drumming measures.
MuseData is presented in Modeling Temporal Dependencies in High-Dimensional Sequences: Application to Polyphonic Music Generation and Transcription by Boulanger-Lewandowski et al.
CCARH's MuseData is a digital library of orchestral and piano classical music. It contains approximately 3 MB of 783 files.
The dataset consists of 305,979 musical notes, each with its pitch, timbre, and envelope. This dataset is likely the most significant number of instrumental notes ever recorded. The musical notes were collected from 1,006 instruments from commercial sample libraries and annotated according to instrument family (acoustic, electronic, or synthetic) and sonic qualities. This dataset includes instruments such as bass, flute guitar, keyboard, mallet, organ, reed, string, synth lead, and vocal.
Over 200 hours of paired audio and MIDI recordings from the International Piano-e-Competition over the last ten years are available in the MAESTRO (MIDI and Audio Edited for Synchronous Tracks and Organisation) dataset. The MIDI data includes key strike velocities and sustain/sostenuto/una corda pedal positions.
Boulanger-Lewandowski et al. introduce JSB Chorales in Modeling Temporal Dependencies in High-Dimensional Sequences: Application to Polyphonic Music Generation and Transcription.
The JSB chorales are a collection of short, four-voice pieces that are notable for their stylistic consistency. Johann Sebastian Bach originally composed the chorales in the 18th century. He composed them by combining pre-existing melodies from contemporary Lutheran hymns and harmonizing them to create parts for the remaining three voices. The dataset version used in representation learning contexts comprises 382 chorales, with a train/validation/test split of 229, 76, and 77 samples, respectively.
URMP (University of Rochester Multi-Modal Musical Performance)
URMP debuted a dataset to aid in the audio-visual analysis of musical performances. The dataset contains several simple multi-instrument musical pieces that researchers created by combining separately recorded performances of individual tracks. A musical score in MIDI format is provided for each piece.
The dataset includes 21.6 million harmonizations from the Bach Doodle and metadata about the composition, such as country of origin and feedback. It also has MIDI of the user-entered melody and MIDI of the generated harmonization. An examination of the melodies in the dataset reveals the most frequently repeated pieces from each country or regional hits.
