!git clone --depth 1 https://github.com/asigalov61/Tegridy-MIDI-Dataset
The Tegridy MIDI Dataset is a comprehensive, multi-instrumental symbolic music resource specifically designed for Music Information Retrieval (MIR) and the development of Music AI models. This repository provides a curated collection of high-quality MIDI files alongside a meticulously assembled ecosystem of external datasets, software tools, and AI applications. The core dataset offerings include original, normalized MIDI collections such as the Tegridy MIDI Dataset, Tegridy Piano, Tegridy Piano-Violin, and Tegridy Children Songs, among others, all released under the CC BY-NC-SA license. Beyond its own data, the project functions as a central hub, cataloging and linking to a vast array of essential external resources. These include massive, processed corpora like Discover (over 6.74 million MIDI files) and Godzilla (over 5.8 million MIDI files), specialized research datasets such as MAESTRO, GiantMIDI, and Groove, and numerous thematic collections. The repository also integrates must-have software (e.g., FluidSynth, MIDI editors), soundfonts for audio rendering, and a wide range of online and Colab-based AI tools for tasks including audio-to-MIDI transcription, MIDI captioning, generation, mixing, and humanization. By providing this unified platform—combining original data, external references, and practical tooling—the Tegridy MIDI Dataset aims to lower the barrier to entry and accelerate research in symbolic music processing, multi-instrumental music generation, and multimodal music understanding.
This list contains a wide variety of MIDI datasets, from massive, processed corpora designed for machine learning to smaller, thematic collections for specific research or hobbies. The descriptions below are based on the content of each provided link. You'll notice a few different "types" of datasets:
- Massive, Curated Datasets (e.g., Discover, Godzilla, Meta): These are enormous collections (often millions of files) that have been aggregated, de-duplicated, normalized, and enriched with extensive metadata and pre-computed features. They are typically designed for large-scale Music Information Retrieval (MIR) and training symbolic music AI models, and often come with their own tooling for search and analysis.
- Specialized Research Datasets (e.g., MAESTRO, GiantMIDI, POP909): These are smaller, high-quality datasets created for specific research tasks. They often feature precise alignments (e.g., note to performed audio), expert annotations (e.g., emotion, structure), or a focus on a particular genre or instrumentation.
- Thematic Collections & Archives (e.g., Touhou MIDI Collection, wild-west-midi-2, classical-music-midi): These are generally smaller, community-driven collections focused on a specific genre, game, or artist. They are often shared on GitHub or Hugging Face and are valuable for fans, hobbyists, and specific analytical tasks.
A massive, production-ready dataset containing over 6.74 million unique, de-duplicated, and normalized MIDI files. It is specifically built for large-scale Music Information Retrieval (MIR), music discovery, and symbolic music AI. The dataset includes rich, precomputed metadata (features counts, genre labels, artist/title IDs) and a custom GPU-accelerated search engine for rapid similarity searches and filtering across the entire collection. It also comes with supplemental code for tasks like loop extraction.
An enormous, comprehensive MIDI dataset with over 5.8 million unique, de-duped, and normalized MIDIs. Each file has been converted to a proper MIDI specification and integrity-checked. It features extensive metadata, including basic features, mono-melody information, pitches-patches counts, and text captions describing the music in each file. Like the Discover dataset, it includes highly optimized, GPU-accelerated search and filter code, making it suitable for large-scale MIR and AI model training.
A MIDI dataset from BreadAi. The dataset contains 5M+ raw deduplicated MIDIs.
A large-scale MIDI dataset repository hosted on GitHub by asigalov61 (the creator of the Los Angeles and Discover datasets). It is one of the foundational collections used in building the larger, more processed datasets. It contains a vast number of MIDI files gathered from various sources on the web.
A dataset from the Metacreation Lab at Simon Fraser University. It is a large collection of MIDI files designed to support research in music generation and MIR. The dataset is notable for its scale and its focus on providing a diverse range of multi-track, multi-instrumental music.
A large-scale dataset of MIDI files specifically curated for training and evaluating machine learning models for music generation. It emphasizes metadata and organization to facilitate research into meta-learning and other advanced AI techniques for music.
A foundational, comprehensive MIDI dataset by Alex Lev (asigalov61). It is a large, multi-instrumental dataset that serves as a basis for many subsequent projects (like Discover and Godzilla). It emphasizes normalization and is designed for a wide range of MIR and music AI tasks.
Lyrics SOTA MIDI dataset by Alex Lev featuring ~179k original MIDI files with matched lyrics, summaries, embeddings, keywords, and chords
The Lakh MIDI Dataset is a well-established and widely-used collection of over 170,000 unique MIDI files. It is particularly famous for its alignment with the Million Song Dataset, allowing for cross-referencing of symbolic data with audio features and metadata. It's a cornerstone dataset for many MIR and generative music projects.
A dataset of piano MIDI files intended for symbolic music modeling. As cited in the Discover/Godzilla pages, it was presented at ICLR 2025 by Louis Bradshaw and Simon Colton. It is likely a high-quality, curated collection focused on piano music for tasks like melody generation or accompaniment.
A dataset that provides a large collection of aligned polyphonic MIDI and lyrics. This makes it particularly valuable for research in singing voice synthesis, lyrics-to-melody alignment, and other tasks that require a direct relationship between notes and text.
A dataset from the XMusic project. Based on the repository name, it is likely designed for research into explainable or controllable music generation, potentially providing rich annotations or structural information alongside the MIDI data.
The Aligned Tracks with Expressive Performance (ATEPP) dataset. It focuses on expressive piano performance, providing MIDI data that captures the nuances of human playing, such as timing and velocity variations, which are often lost in quantized MIDI files.
A large-scale, high-quality piano MIDI dataset from ByteDance. It contains over 10,000 solo piano pieces transcribed from audio recordings, covering a wide range of classical and other genres. It is a key resource for piano-related AI tasks.
A dataset focused on classical music MIDI files. It provides a collection suitable for training models on classical forms, composers, and styles.
A specialized dataset by asigalov61 focusing on piano performances in the rock genre. It's a valuable resource for studying and generating rock music from a keyboard-centric perspective.
The MAESTRO (MIDI and Audio Edited for Synchronous Tracks and Organization) dataset is a premier resource for piano research. It consists of over 200 hours of virtuosic piano performances, captured via a Yamaha Disklavier, which provides precisely aligned audio and MIDI data. It is essential for tasks like automatic music transcription and performance analysis.
A dataset of piano MIDI files, likely sourced from the internet and organized for research. It provides a broad collection of piano music in a symbolic format.
A dataset of 909 popular songs with professional, multi-track annotations. It includes separate tracks for melody, accompaniment, and the main vocal, as well as tempo and key annotations. It is highly valuable for research in popular music structure, arrangement, and performance.
A large-scale dataset of popular music, containing over 1,700 songs with professionally transcribed MIDI files. It likely provides rich annotations for chords, melody, and structure, serving as a significant resource for popular music analysis.
A dataset designed for piano jazz solo transcription and analysis. It provides high-quality MIDI alignments of jazz piano performances, which is crucial for research into jazz harmony, improvisation, and swing feel.
A classic, smaller dataset of over 1,000 British and American folk tunes. It is often used as a benchmark for sequence modeling and melody generation due to its simple, homophonic structure and clear melodic lines.
The Aligned Scores and Performances (ASAP) dataset. It bundles 2229 piano performances (MIDI) aligned with 284 musical scores (in symbolic format). This alignment is a powerful feature for studying expressive performance, score following, and music synchronization.
A dataset specifically for lead sheets, which contain the melody, harmony (chords), and lyrics of a song. It's an important resource for research in harmonic analysis, melody-chord relationships, and automatic lead sheet generation.
A dataset likely focused on cross-cultural music analysis, possibly comparing elements of Western and Eastern (e.g., Chinese) music. The name suggests a duality, which could refer to different musical attributes or cultural origins.
The Meertens Tune Collections (MTC) is a dataset of Dutch folk songs. It provides a rich, curated collection of melodies with associated metadata, useful for ethnomusicological studies and folk music generation.
This is not a single dataset but a massive, online library of musical scores in the **Humdrum kern format. It contains thousands of high-quality, scholarly encoded scores, primarily of classical and early music. It is an invaluable resource for music theory and computational musicology.
A dataset of pop piano music with emotion annotations. It provides MIDI files of pop piano performances, each labeled with a quadrant from the arousal-valence emotion model, enabling research into emotion recognition and emotion-conditioned music generation.
A dataset focused on the relationship between chords and melody. It provides paired data, which is ideal for training models for tasks like melody harmonization or chord-conditioned melody generation.
The Groove MIDI Dataset is a large-scale, high-quality dataset of drum performances. It contains over 13 hours of recorded drumming with detailed annotations, making it the go-to resource for modeling, generating, and analyzing drum patterns and "groove."
An expansion of the original Groove MIDI Dataset, adding even more performances, drummers, and stylistic variety. It further solidifies the Groove dataset as the primary resource for data-driven drum research.
A dataset curated by asigalov61 that provides MIDI files with added annotations. This likely includes metadata such as key, tempo, chord progressions, or structural boundaries, making it useful for supervised learning tasks.
A dataset specifically created for the task of aligning music (MIDI) with corresponding text. This is a key resource for research in lyrics transcription, lyrics-to-audio alignment, and multimodal music understanding.
A dataset designed to link MIDI files with textual descriptions (captions) . It pairs symbolic music with natural language, which is essential for developing text-to-music retrieval systems and multimodal generative models.
A dataset focusing on monophonic MIDI and the task of transposition. It can be used to train or evaluate models that can transpose melodies or monophonic lines to different keys while maintaining musicality.
A dataset for conditional music generation. It provides MIDI data along with a rich set of performance instructions (e.g., genre, tempo, instruments, mood), allowing for fine-grained control over the output of generative models.
A dataset likely focused on fine-grained music editing or inpainting. "FiLD" could stand for Fill In the middle or a similar concept, providing data and tasks for models that can intelligently insert or replace musical sections within a piece.
A dataset designed specifically for the task of identifying which track in a multi-track MIDI file carries the main melody. This is a crucial pre-processing step for many MIR and generative music systems.
The dataset associated with the SymphonyNet model for symphonic music generation. It contains a large collection of multi-instrumental, orchestral MIDI files, enabling research into large-scale, structured music generation.
A small, specific dataset likely containing MIDI data from Casio keyboards or related to Casio's music data format. It might be useful for hardware interoperability or analyzing preset music from these devices.
A dataset focused on identifying and extracting "hooks"—the catchy, memorable parts of popular songs. It provides MIDI data with annotations marking these sections, which is valuable for music information retrieval and analysis of popular music structure.
A collection of piano MIDI files, likely sourced from public online repositories ("pub" as in public). It provides a broad, uncurated set of piano music for various tasks.
The Automated Counterpoint and Polyphonic Arrangement Synthesis dataset. It is likely designed for research into counterpoint and polyphonic writing, providing examples of well-formed polyphonic music for training generative models.
A dataset likely related to synthesizer or electronic music. The name suggests it contains synthesized music or is designed for use with synthesizers, possibly including parameter settings alongside MIDI data.
The Salzburg MIDI Dataset (SMD). It is a collection of MIDI files, likely with a focus on classical music, curated by a research institution. The Zenodo link points to a specific version of this dataset.
A dataset for the task of orchestration—taking a piano reduction and expanding it into a full orchestral arrangement. It likely contains pairs of piano scores and their corresponding multi-track orchestral MIDI versions.
Another classical music MIDI collection on Hugging Face. It serves as a readily accessible source of classical music in symbolic format for various machine learning and analysis projects.
A dataset with the acronym "GAPS". Without further context, it is a specialized collection hosted on Zenodo, likely for a specific MIR task. The name might hint at its content, such as "Guitar and Piano Scores" or "Genre-Annotated Polyphonic Set".
The Bach Piano Scores Dataset (BPSD). A specialized collection of J.S. Bach's piano works encoded in MIDI, providing a high-quality, focused resource for studying and modeling Baroque counterpoint and structure.
A repository from Nightingale AI containing MIDI data. It likely serves as a resource for their internal models or as a general collection for the AI music community.
A personal, hobbyist collection of MIDI files shared on GitHub. It likely contains arrangements and transcriptions gathered by the user, possibly with a focus on video game or chiptune music.
A general-purpose archive of MIDI files. As a user-maintained archive, it aims to collect and preserve a wide variety of MIDI files from across the internet.
A GitHub repository likely containing MIDI files, possibly related to a specific project or game. "Moon Antonio" could be a username or project name.
A project repository from a university course (CSCI599). It contains the dataset and code used for a project applying Transformers to music generation or analysis.
A small repository of MIDI files intended for demonstration or testing purposes, likely for software development or simple examples.
A project repository for a music generation model. It likely includes a dataset of MIDI files used to train the model, along with the generation code.
A repository for a music generation project. It contains the dataset and code used by the author for their work on algorithmic music composition.
A university course repository (MC906) that likely includes a MIDI dataset component for a project on AI and music, alongside other AI topics.
Another general-purpose, user-maintained archive of MIDI files collected from various online sources.
A repository likely containing MIDI files related to the World of Warcraft game. These are probably transcriptions of the game's soundtrack or user-created arrangements for the in-game music system.
A personal repository containing various data projects, which may include a small collection of MIDI files as part of one of those projects.
A personal collection of MIDI files shared by a user named "monsterjd95". The name suggests a casual, hobbyist collection.
A user repository that likely contains a collection of MIDI files, possibly for a game modding project or a personal music archive.
A repository for a project called "Midimancy Hubworld." It likely contains the MIDI files associated with that project, which may be a game, interactive experience, or music tool.
A repository dedicated to piano MIDI files, possibly created or arranged by a user named Trentorial.
A GitHub repository containing a collection of MIDI files, primarily focused on rock and alternative music (e.g., Foo Fighters, Linkin Park, Radiohead). It appears to be a user's personal archive of song transcriptions or arrangements.
A Hugging Face dataset of classical music MIDI files. The file viewer shows a list of files with classical composer-style names (e.g., "ballade2", "beethoveenrondo"), suggesting a collection of classical piano works.
A comprehensive, community-driven collection of MIDI files from the TouHou Project game series and its related music. It is meticulously organized by game era (shooters, fighting games, music CDs) and includes fan arrangements. The repository provides detailed information on the original sound hardware (Roland Sound Canvas) used by the composer ZUN, which is crucial for achieving authentic playback.
This Internet Archive page hosts "The Nuker Series," a collection of Black MIDI files. Black MIDIs are characterized by an extremely high number of notes, often pushing the limits of the MIDI format.
The datasets below are a crucial complement to the previous MIDI-only datasets. They provide paired audio and MIDI data, which is essential for tasks like automatic music transcription (AMT), source separation, and performance analysis. They range from solo piano to multi-instrument ensembles.
A collection of 330 freely-licensed classical music recordings (primarily chamber music) with over 1 million annotated labels. It provides precise time-aligned annotations for each note, including the instrument playing it and its position in the metrical structure. Labels were acquired by aligning musical scores to recordings and verified by trained musicians, with an estimated error rate of just 4%. The dataset includes the audio (.wav), label files (.csv), and the reference MIDI files used to construct the annotations, making it a robust benchmark for supervised music transcription.
The MAESTRO (MIDI and Audio Edited for Synchronous TRacks and Organization) dataset is a premier resource consisting of about 200 hours of virtuosic piano performances captured on Yamaha Disklaviers. This setup provides high-precision alignment (~3 ms accuracy) between the audio waveforms and the MIDI data, which includes key strike velocities and pedal positions. Sourced from ten years of the International Piano-e-Competition, the repertoire is mostly classical. It comes with a proposed train/validation/test split designed to ensure the same composition does not appear in multiple subsets.
A dataset specifically designed for guitar transcription, providing high-quality recordings alongside rich, time-aligned annotations. It contains 360 excerpts (approx. 30 seconds each) played by six different guitarists across five musical styles. A key feature is the use of a hexaphonic pickup, which records each string separately, enabling largely automated and precise annotation. Data for each excerpt includes the hexaphonic audio, a reference microphone recording, and a JAMS file with annotations for pitch contours per string, MIDI notes, chords, beats, and playing style.
The University of Rochester Multi-Modal Music Performance (URMP) dataset focuses on multi-instrument classical music. It comprises 44 simple multi-instrument pieces assembled from coordinated but separately recorded individual tracks. For each piece, it provides the musical score (MIDI), the high-quality individual audio recordings for each instrument, and a video of the assembled performance. This unique multi-modal approach makes it highly valuable for tasks like music source separation, multi-instrument transcription, and audio-visual performance analysis.
The Synthesized Lakh (Slakh) Dataset is a large-scale dataset for music source separation and multi-instrument automatic transcription. It contains 2100 automatically mixed multi-track audio pieces, synthesized from the Lakh MIDI Dataset using professional, sample-based virtual instruments. The result is 145 hours of mixture data with perfectly aligned, ground-truth MIDI files for each individual instrument track (e.g., Piano, Guitar, Drums, Bass). All audio is provided in FLAC format, and the dataset includes detailed metadata for each source track.
A curated list of online demos, colab notebooks, and software for audio processing and MIDI manipulation.
- [Online] Vocal Separation 2025 - A Hugging Face Space for isolating vocals from a mixed audio track using deep learning models. This new space includes Piano separation with Spleeter 2025.
- [Online] Vocal Separation - A Hugging Face Space for isolating vocals from a mixed audio track using deep learning models.
- [Online] Music Flamingo - An NVIDIA demo that generates natural language descriptions (captions) for music audio clips.
- [Online] Sonic Verse - A tool from AMAAI lab for generating textual descriptions or understanding the narrative within audio scenes.
- [Online] GAMA IT - A demo for generating captions for audio, likely focusing on general audio events and music.
- [Colab] ByteDance Solo Piano with Pedals Audio-to-MIDI Transcription - A Google Colab notebook for transcribing solo piano performances, including pedal information, into MIDI format.
- [Online] ByteDance Solo Piano with Pedals Audio-to-MIDI Transcription - An online Hugging Face demo of the ByteDance model for transcribing solo piano audio to MIDI with pedal data.
- [Online] YourMT3 Multi-Instrumental Audio-to-MIDI Transcription - A demo for transcribing polyphonic, multi-instrumental music from audio to MIDI using a custom implementation of the MT3 architecture.
- [Online] Google MT3 Multi-Instrumental Audio-to-MIDI Transcription - An online implementation of Google's MT3 (Music Transcription Transformer) model for transcribing multiple instruments from audio.
- [Colab] Google Multi-Instrumental and Solo Piano Audio-to-MIDI Transcription - The official Google Colab notebook for Magenta's MT3 model, enabling multi-instrumental and solo piano transcription.
- [Online] Ultimate MIDI Captioner - Ultimate MIDI captioner and analyzer
- [Online] MIDI Music Flamingo - AI model that generates textual descriptions for MIDI files, similar to image or audio captioning but for symbolic music.
- [Online] Advanced MIDI Renderer - A sophisticated online tool for rendering MIDI files to high-quality audio using soundfonts.
- [Online] Audio to MIDI and Advanced Renderer - A combined tool that first transcribes audio to MIDI and then renders the resulting MIDI file to audio.
- [GitHub] fluidsynth - A real-time software synthesizer that reads and renders MIDI files to audio using SoundFont samples. A foundational tool in this space.
- [GitHub] SpessaSynth - A modern, web-based MIDI synthesizer and player written in JavaScript, capable of running in browsers.
- [GitHub] midirenderer - A Python command-line tool for batch rendering MIDI files to WAV audio using FluidSynth.
- [Software] MIDITrail - A 3D MIDI visualization software that displays notes falling down a "trail," providing an immersive view of the music's structure.
- [Software] midis2jam2 - A 3D visualizer that animates instruments and notes from a MIDI file, creating a fun and engaging music video.
- [Software] MIDIVisualizer - A highly customizable, real-time 3D visualization application for MIDI files, built with OpenGL.
- [Software] MIDIFall - A "falling notes" style visualizer (like in Guitar Hero) for MIDI files, built with Unity.
- [GitHub] euphony - A web-based, interactive 3D MIDI visualizer that runs in the browser using Three.js.
- [GitHub] midi picasso - A creative tool that generates static visual art pieces based on the data and structure of a MIDI file.
- [Online] MIDI Identification - A tool that analyzes a MIDI file to identify its key, time signature, tempo, and other musical attributes.
- [Online] Los Angeles MIDI Dataset Search - A search engine for exploring and finding MIDI files within the large Los Angeles MIDI Dataset.
- [Online] LAKH MIDI Dataset Search - A search interface for the extensive Lakh MIDI Dataset, allowing users to find MIDI files by name or other criteria.
- [Online] Advanced MIDI Search - A general-purpose search tool for finding MIDI files based on musical content like chord progressions or note patterns.
- [Online] Karaoke MIDI Search - A specialized search engine for finding MIDI files suitable for karaoke, likely focusing on those with melody and lyric tracks.
- [Online] MIDI Genre Classifier - A tool that predicts the musical genre of a given MIDI file.
- [Online] Ultimate MIDI Classifier - An advanced classifier that likely categorizes MIDI files by multiple attributes, such as genre, mood, or instrumentation.
- [Online] Advanced MIDI Classifier - Another robust tool for classifying MIDI files, potentially with more granular categories or different underlying models.
- [Online] Orpheus MIDI Comparator - A tool from Project Los Angeles for analyzing and quantifying the differences between two MIDI files.
- [Online] Intelligent MIDI Comparator - A smart tool for comparing MIDI files, likely focusing on musical similarities and differences rather than just raw data.
- [Online] MIDI Doctor - An online utility that automatically detects and fixes common issues in MIDI files, such as overlapping notes, incorrect tempos, or corrupted data.
- [Online] MIDI Aligner - A tool for synchronizing a MIDI file with an audio performance, correcting timing offsets to align the two.
- [Online] Orpheus Bridge Music Transformer - An AI model that generates a musical bridge to connect two separate MIDI segments.
- [Online] Orpheus Pitches Inpainter - A tool that fills in missing or "masked" notes within a MIDI sequence (inpainting), harmonically completing the musical phrase.
- [Online] Orpheus MIDI Loops Mixer - A tool for intelligently combining and arranging different MIDI loops to create a coherent musical piece.
- [Online] Orpheus MIDI Mono Melodies Mixer - A tool designed to mix two or more monophonic (single-note) MIDI melodies into a harmonized polyphonic output.
- [Online] MIDI Remixer - A general-purpose remixing tool that can alter the style, instrumentation, or structure of an input MIDI file.
- [Online] MIDI Loops Mixer - A tool for layering and combining multiple MIDI loops, likely with automatic key and tempo matching.
- [Online] MIDI Chords Mixer - A tool for intelligently combining and reharmonizing chord progressions from different MIDI sources.
- [Online] MIDI Splitter - A tool for splitting any MIDI into individual voices parts
- [Online] Orpheus Music Segmentator - An AI model that segments any MIDI into coherent separate parts
- [Online] Orpheus Humanizing Transformer - An AI model that adds expressive timing and velocity variations to a "perfect" but robotic-sounding MIDI file to make it sound more human-performed.
- [Online] Awesome Drums Transformer - An AI model for generating realistic and creative drum patterns in MIDI format.
- [Online] Godzilla Piano Chords Texturing Transformer - A tool that generates accompaniment patterns or "textures" (like arpeggios, broken chords) from a given piano chord progression.
- [Online] MIDI Melody - A simple tool for extracting or generating the primary melody line from a MIDI file.
- [Online] Harmonic Melody MIDI Mixer - A tool for mixing two melodies, ensuring the resulting output is harmonically coherent.
- [Online] Chords Progressions Generator - An AI tool for generating novel chord progressions in various styles and keys.
- [Online] Melody Harmonizer Transformer - A transformer-based model that takes a monophonic melody and generates a harmonized chord progression to accompany it.
- [Online] Mono Melodies Generator - An AI model for generating new monophonic (single-line) melodies from scratch or from a seed.
- [Online] Parsons Code Melody Generator - A tool that generates melodies based on the Parsons Code, a symbolic notation for melodic contours (direction of pitch movement).
- [Online] MuseCraft Chords Progressions - A chord progression generator from the MuseCraft suite, focused on crafting harmonic foundations.
- [Online] MuseCraft AlgoPOP - A generative tool from MuseCraft designed to create pop-music-inspired structures and melodies.
- [Online] Lyrics Morpher - A creative tool for transforming or "morphing" input lyrics into new variations, potentially for songwriting inspiration.
@misc{Lev2026TegridyMIDIDataset,
author = {Alex Lev},
title = {Tegridy MIDI Dataset: A Comprehensive Multi-Instrumental MIDI Resource for Music AI and MIR Research},
year = {2026},
howpublished = {\url{https://github.com/asigalov61/Tegridy-MIDI-Dataset}},
organization = {GitHub},
note = {Includes original Tegridy collections, curated links to major external datasets (Discover, Godzilla, MAESTRO, Lakh, etc.), software tools, and AI applications. Licensed under CC BY-NC-SA 4.0.}
}