Frequently asked questions

List of frequently asked questions

• How can I collaborate or get in touch?

  I work alone and I read everything that arrives. If you have an idea, a question about the research or want to propose some kind of collaboration, email is the direct route: [email protected]. I have no contact form or ticketing system; a real conversation starts with a real message.

  To follow the academic work in the researcher registry, my ORCID identifier is 0009-0004-9691-3396. That is where affiliations and output are recorded. The code for the projects — tools, experiments, analysis scripts — is on GitHub under the profile musicTever. Not everything is published at the moment of writing, but what is there is functional or explicitly marked as a prototype.

  You can find all contact channels and profiles on Contact.

• What does your doctoral research involve?

  My doctoral research focuses on automatic polyphonic music transcription: pieces where several notes — and often several voices — sound at once, the hard case within AMT. I work on piano, taking current state-of-the-art systems as a starting point — Basic Pitch, Onsets & Frames, MT3 among them —, analysing where and why they fail note by note, not just their overall accuracy score. From there I explore alternative audio representations that might correct some of those systematic failures.

  The gaita asturiana repertoire is not my current corpus, but it’s part of why this research matters to me: acoustically it’s a complex instrument — the chanter and the drone sound continuously and simultaneously, with non-equal-tempered tuning —, and it’s underrepresented in international MIR benchmarks.

  I am a pre-doctoral candidate in the 2026/27 academic year. The work I publish here reflects the current state of the research: ongoing hypotheses, tested tools, preliminary results. You can see the active and closed projects on Projects.

• What is Automatic Music Transcription (AMT)?

  Automatic Music Transcription (AMT) is the area of music informatics concerned with converting an audio signal into a symbolic representation: notes with their pitches, durations and intensities, as if an expert listener were writing down in score notation what they hear. The challenge is non-trivial: an audio signal mixes sources, noise, vibrato and tempo variations that no simple formula can disentangle.

  AMT algorithms work in several layers. First they extract spectral features from the signal — the Fourier transform is the usual starting point, as I explain in The Fourier Transform — then they detect events (note onset, duration, pitch), and finally they structure those events into a usable representation: MIDI, MusicXML or a proprietary format.

  My research applies AMT to polyphonic piano music, taking current state-of-the-art systems as a starting point — Basic Pitch, Onsets & Frames, MT3 — and analysing where and why they fail note by note. Instruments such as the gaita asturiana, with intrinsic polyphony (chanter and drone always sound simultaneously) and non-equal temperament, are a separate research interest: their characteristics would complicate generic models even further, but they are not my current working corpus.

• What is labs.tever.es?

  labs.tever.es is the technical and research space of Tever Díaz García: an open research notebook on music informatics, Automatic Music Transcription (AMT), Music Information Retrieval (MIR) and digital humanities. I publish the process here — prototypes, tools, methodological reflections — of research currently focused on automatic transcription of polyphonic piano music.

  This site is the technical sibling of tever.es, where the musical side lives: discography, performances and teaching. The division is deliberate: here you will find code, signals and algorithms; there you will find chanters and drones. Two sites, one person.

  The format follows the open notebook model (open notebook science): I do not wait for definitive results before publishing. What appears here may be a half-built prototype, a preliminary analysis or an unanswered question. The transparency of the process is part of the project.

  You can read more about who is behind this notebook on About.

• What is Music Information Retrieval (MIR)?

  Music Information Retrieval (MIR) is the field of computer science that studies how to extract, organise and retrieve information from musical signals. It covers a wide range of tasks: chord recognition, source separation (isolating a voice from a guitar in a mix), tempo identification, genre classification, audio fingerprint-based retrieval, and — the central task for my work — automatic music transcription. If you have ever seen Shazam identify a song in three seconds, you are watching applied MIR.

  A necessary clarification: in Spanish-speaking contexts, the acronym MIR is also used for Médico Interno Residente (medical residency). The MIR discussed here has nothing to do with medical specialisation; it is a discipline within computer science and computational sciences, with its own research community, its own conferences (ISMIR) and its own reference corpora.

  My research is situated within MIR applied to automatic music transcription (AMT): I work on polyphonic piano, analysing where and why current state-of-the-art systems fail. The gaita asturiana repertoire — underrepresented in international MIR benchmarks, which have historically been biased towards Western pop music in equal temperament — is a separate research interest, not my current working corpus.

• What is the MIDI gaita and what does it have to do with the research?

  The Gaita Hero project was my undergraduate final-year project in 2011: building a gaita asturiana that generated MIDI events from the player’s gestures. The idea was to capture finger movements on the chanter — the fingering — and translate them into standard MIDI messages that any synthesiser could interpret. It was, in essence, the first prototype of a gaita asturiana as a digital instrument.

  That work explored a deeper question that still interests me: how to capture the musical gesture of an oral-tradition instrument (with its non-equal temperament, its specific technique and its intrinsic polyphony) and how to represent it in formats that computing systems can process. Although my current working corpus in automatic music transcription is piano, the MIDI gaita was my first contact with that underlying question: before transcribing audio, you need to understand what it is you are transcribing.

  In 2021, that line of work had a creative continuation with the video game of the same name, which used the mechanisms of the original project in a playful context. You can see the details of both projects on Projects.

• Why do you publish your research in the open?

  This notebook follows the open notebook science model: publishing the research process as it happens, not only when there is a validated result to present. That includes half-built prototypes, preliminary analyses, experiments that did not work as expected and questions that still have no answer. The transparency of the process is part of the project, not a side effect.

  The reason is straightforward: research is not linear. Between the initial hypothesis and a publishable result there is a territory of real work — methodological decisions, dead ends, shifts in approach — that normally disappears from the academic record. Publishing that territory is valuable for anyone working on similar problems, and it also obliges me to be precise about what I know and what I do not.

  The boundaries are clear: anything unverified is marked as such, conjectures are never presented as results, and no credentials or publications are invented. That honesty is the foundation of the Digital Bond Manifesto, which articulates the underlying philosophy behind how I publish in the open.

• What is MusicXML?

  MusicXML is the standard format for two different music programs to exchange a score without losing information. It is maintained by the W3C Music Notation Community Group, is at version 4.0 (2021), and is supported by more than 250 notation, editing and analysis applications.

  Unlike MIDI —which describes a performance: which note sounds, when, and how hard—, MusicXML describes the written score: clefs, key signatures, slurs, articulations and lyrics. That is why it is readable XML text rather than a binary: each musical element is stored as structured data, which makes it possible to edit, print, archive openly and analyse it. I explain it in detail in What MusicXML is.