Datasets

Crowdsourced low- and high-level acoustic descriptors indexed by MusicBrainz recording identifiers.

Four genre-annotation datasets joining AcousticBrainz features with AllMusic, Discogs, Last.fm and Tagtraum labels.

Crowdsourced natural-language captions paired with AudioSet-derived audio clips.

Human-labeled ten-second YouTube segments spanning music, speech and environmental sound classes.

User-entered melodies and model-generated four-voice harmonizations contributed through the 2019 Bach Doodle.

A test-only benchmark converting diverse MIR tasks into standardized music instruction-following evaluation.

Preference data and evaluation for reward models scoring musicality, text alignment and compositional instruction following.

Popular-music signals with lyrics and listener-derived intelligibility scores for the ICASSP 2026 Cadenza Challenge.

Synthetic four-part chamber-ensemble music with mixtures, stems, MIDI and fine-grained performance annotations.

Controlled singing-voice deepfake corpus generated from public singing data with synthesis and voice-conversion systems.

Full-song lyric, vocal-note and audio alignments at note, word, line and paragraph granularity.

Music excerpts and full songs with continuous and song-level valence and arousal annotations.

Pop-piano audio and MIDI clips labeled for perceived emotion, supporting recognition and emotion-conditioned generation.

Expanded human drum-performance corpus with MIDI-aligned audio rendered across 43 drum kits.

Open human-labeled Freesound clips across 200 AudioSet-ontology classes, including many musical sounds.

MusicCaps prompts regenerated with multiple text-to-music systems for synthetic-music detection and generator attribution.

Full-length Creative Commons music with audio, features, metadata and hierarchical genre labels for MIR.

Classical-guitar audio, scores and high-resolution MIDI alignments for transcription research.

Machine-transcribed solo-piano MIDI derived from web recordings and IMSLP work metadata.

Human-performed expressive drum MIDI with aligned synthesized audio and performance metadata.

Acoustic-guitar recordings with hexaphonic audio and time-aligned note, chord, beat and technique annotations.

A NeurIPS challenge and evaluation suite for general-purpose audio representations across speech, sound and music tasks.

A completed grand challenge for predicting overall and five fine-grained human aesthetic ratings of AI-generated songs.

A completed challenge to predict lyric intelligibility for popular music, including listeners with hearing loss.

Large deduplicated MIDI collection with matched and aligned subsets linked to the Million Song Dataset.

Multi-generator, multi-domain, multilingual, multicultural, multi-genre and multi-instrument music-detection databases.

Large music-aesthetics benchmark with multi-dimensional professional ratings and textual comments.

Paired piano audio and high-precision MIDI from International Piano-e-Competition performances.

A unified evaluation suite for pretrained music-audio representations across acoustic, performance, score and high-level tasks.

A MIREX task evaluating systems that generate natural-language descriptions for music audio.

A multi-domain benchmark of expert audio understanding and reasoning spanning speech, environmental sound and music.

Expert-created audio questions covering speech, sound, music, mixtures, spatial audio and long-form reasoning.

Full-length stereo music mixtures with isolated drums, bass, vocals and other stems for source separation.

Magnatune audio clips with human tags collected through the TagATune game for music auto-tagging research.

Chord, structure and feature annotations for a sampled set of Billboard-chart songs.

Annotated royalty-free multitrack recordings supporting melody, pitch, instrument and source-separation research.

Audio features and metadata for one million contemporary popular-music tracks, without distributed audio.

Human-validated multiple-choice questions for evaluating music understanding and reasoning in audio-language models.

A live pairwise evaluation platform and rolling leaderboard for text-to-music generation systems.

Rolling human preference data from live pairwise comparisons of text-to-music model outputs.

Multimodal music features and large-scale Last.fm listening histories extending Music4All.

Music audio-text pairs expanding MusicCaps with extracted musical features and augmentation for controllable generation.

Dataset of 5.5k music clips with expert-written text captions

Classical recordings aligned to score-derived note, instrument and metrical-position labels.

Annotated four-second musical-note audio for timbre modeling, synthesis, and audio representation research.

Creative Commons music excerpts partially labeled for presence or absence of 20 instrument classes.

Large MusicXML collection of scores labeled public domain, with score and user-interaction metadata.

Professional piano arrangements of 909 popular songs with aligned MIDI and structural annotations.

Hierarchical expert structural annotations for a large and stylistically diverse music collection.

The inaugural controlled and in-the-wild singing-voice deepfake detection challenge held with IEEE SLT 2024.

In-the-wild real and deepfake singing data for singing-voice authenticity detection.

Synthetic multitrack audio and aligned MIDI rendered from Lakh MIDI files for separation and transcription.

Crowdsourced human descriptions of Creative Commons music for music-language evaluation.

Full-length real and generated songs with expert ratings across five aesthetic dimensions.

A completed AIcrowd challenge with music and cinematic source-separation tracks and hidden evaluation sets.

One million anonymized public playlists released for automatic playlist continuation research.

Metadata and downloadable links for 659,788 Suno-generated songs discovered through systematic search queries.

Coordinated multi-instrument classical performances with separated audio, assembled video, scores and pitch annotations.

A cappella singing recordings across singers, vowels, registers and vocal techniques, with a corrected annotated derivative.

Large weakly labeled audio-caption corpus assembled from web sound libraries and AudioSet, with ChatGPT-assisted caption cleaning.

In-the-wild singing voice deepfake detection data extending SingFake for the SVDD 2024 challenge.