48 machine learning datasets
48 dataset results
MUSAN is a corpus of music, speech and noise. This dataset is suitable for training models for voice activity detection (VAD) and music/speech discrimination. The dataset consists of music from several genres, speech from twelve languages, and a wide assortment of technical and non-technical noises.
The MAESTRO dataset contains over 200 hours of paired audio and MIDI recordings from ten years of International Piano-e-Competition. The MIDI data includes key strike velocities and sustain/sostenuto/una corda pedal positions. Audio and MIDI files are aligned with ∼3 ms accuracy and sliced to individual musical pieces, which are annotated with composer, title, and year of performance. Uncompressed audio is of CD quality or higher (44.1–48 kHz 16-bit PCM stereo).
MusicCaps is a dataset composed of 5.5k music-text pairs, with rich text descriptions provided by human experts. For each 10-second music clip, MusicCaps provides:
MusicNet is a collection of 330 freely-licensed classical music recordings, together with over 1 million annotated labels indicating the precise time of each note in every recording, the instrument that plays each note, and the note's position in the metrical structure of the composition. The labels are acquired from musical scores aligned to recordings by dynamic time warping. The labels are verified by trained musicians; we estimate a labeling error rate of 4%. We offer the MusicNet labels to the machine learning and music communities as a resource for training models and a common benchmark for comparing results.
The MTG-Jamendo dataset is an open dataset for music auto-tagging. The dataset contains over 55,000 full audio tracks with 195 tags categories (87 genre tags, 40 instrument tags, and 56 mood/theme tags). It is built using music available at Jamendo under Creative Commons licenses and tags provided by content uploaders. All audio is distributed in 320kbps MP3 format.
The JSB chorales are a set of short, four-voice pieces of music well-noted for their stylistic homogeneity. The chorales were originally composed by Johann Sebastian Bach in the 18th century. He wrote them by first taking pre-existing melodies from contemporary Lutheran hymns and then harmonising them to create the parts for the remaining three voices. The version of the dataset used canonically in representation learning contexts consists of 382 such chorales, with a train/validation/test split of 229, 76 and 77 samples respectively.
Click to add a brief description of the dataset (Markdown and LaTeX enabled).
ASAP is a dataset of 222 digital musical scores aligned with 1068 performances (more than 92 hours) of Western classical piano music.
The Lakh Pianoroll Dataset (LPD) is a collection of 174,154 multitrack pianorolls derived from the Lakh MIDI Dataset (LMD).
First large-scale symphony generation dataset.
ATEPP is a dataset of expressive piano performances by virtuoso pianists. The dataset contains 11677 performances (~1000 hours) by 49 pianists and covers 1580 movements by 25 composers. All of the MIDI files in the dataset come from the piano transcription of existing audio recordings of piano performances. Scores in MusicXML format are also available for around half of the tracks. The dataset is organized and aligned by compositions and movements for comparative studies.
The rise of singing voice synthesis presents critical challenges to artists and industry stakeholders over unauthorized voice usage. Unlike synthesized speech, synthesized singing voices are typically released in songs containing strong background music that may hide synthesis artifacts. Additionally, singing voices present different acoustic and linguistic characteristics from speech utterances. These unique properties make singing voice deepfake detection a relevant but significantly different problem from synthetic speech detection. In this work, we propose the singing voice deepfake detection task. We first present SingFake, the first curated in-the-wild dataset consisting of 28.93 hours of bonafide and 29.40 hours of deepfake song clips in five languages from 40 singers. We provide a train/val/test split where the test sets include various scenarios. We then use SingFake to evaluate four state-of-the-art speech countermeasure systems trained on speech utterances. We find these sys
AIOZ-GDANCE comprises 16.7 hours of whole-body motion and music audio of group dancing. The duration of each video in our dataset is ranging from 15 to 60 seconds.
The MusicBench dataset is a music audio-text pair dataset that was designed for text-to-music generation purpose and released along with Mustango text-to-music model. MusicBench is based on the MusicCaps dataset, which it expands from 5,521 samples to 52,768 training and 400 test samples!
The scarcity of high-quality and multi-task singing datasets significantly hinders the development of diverse controllable and personalized singing tasks, as existing singing datasets suffer from low quality, limited diversity of languages and singers, absence of multi-technique information and realistic music scores, and poor task suitability. To tackle these problems, we present GTSinger, a large Global, multi-Technique, free-to-use, high-quality singing corpus with realistic music scores, designed for all singing tasks, along with its benchmarks. Particularly, (1) we collect 80.59 hours of high-quality songs, forming the largest recorded singing dataset; (2) 20 professional singers across nine languages offer diverse timbres and styles; (3) we provide controlled comparison and phoneme-level annotations of six singing techniques, helping technique modeling and control; (4) GTSinger offers realistic music scores, assisting real-world musical composition; (5) singing voices are accompa
MuChoMusic is a benchmark designed to evaluate music understanding in multimodal language models focused on audio. It includes 1,187 multiple-choice questions validated by human annotators, based on 644 music tracks from two publicly available music datasets. These questions cover a wide variety of genres and assess knowledge and reasoning across several musical concepts and their cultural and functional contexts. The benchmark provides a holistic evaluation of five open-source models, revealing challenges such as over-reliance on the language modality and highlighting the need for better multimodal integration.
The Song Describer Dataset (SDD) contains ~1.1k captions for 706 permissively licensed music recordings. It is designed for use in evaluation of models that address music-and-language (M&L) tasks such as music captioning, text-to-music generation and music-language retrieval.
A large-scale and diverse duet interactive dance dataset. Recording about 117 minutes of professional dancers' performances.
ComMU has 11,144 MIDI samples that consist of short note sequences created by professional composers with their corresponding 12 metadata. This dataset is designed for a new task, combinatorial music generation which generate diverse and high-quality music only with metadata through auto-regressive language model.
For each dataset we provide a short description as well as some characterization metrics. It includes the number of instances (m), number of attributes (d), number of labels (q), cardinality (Card), density (Dens), diversity (Div), average Imbalance Ratio per label (avgIR), ratio of unconditionally dependent label pairs by chi-square test (rDep) and complexity, defined as m × q × d as in [Read 2010]. Cardinality measures the average number of labels associated with each instance, and density is defined as cardinality divided by the number of labels. Diversity represents the percentage of labelsets present in the dataset divided by the number of possible labelsets. The avgIR measures the average degree of imbalance of all labels, the greater avgIR, the greater the imbalance of the dataset. Finally, rDep measures the proportion of pairs of labels that are dependent at 99% confidence. A broader description of all the characterization metrics and the used partition methods are described in