Essay · DIY Lab
Gaita Hero and EMTI Fervienza: evaluating the sound
In 2011 I designed a system to check Asturian bagpipe scores. I'm still pulling the same thread: comparing a performance against a reference is an MIR problem.
In January 2011, La Nueva España covered my final-year university project with a headline that still makes me smile: Gaita Hero, an obvious nod to Guitar Hero. The name wasn’t mine — it was the press’s — but the underlying question the project asked was mine, and I’ve been chasing it ever since. Can a computer listen to how you play and tell you whether you’ve played what you were supposed to play correctly?
The project behind the newspaper name
In 2011, at the School of Computer Engineering of the University of Oviedo, I presented a final-year project co-supervised with Balbino Menéndez Suárez, master piper and director of the Camín de Fierro band (Proaza, Santo Adriano, and Teverga). The idea was more modest than the headline suggests: we used a real Asturian bagpipe to check whether a student played simple pieces correctly. The system proposed one of those scores, the student played it in front of a microphone, and the program checked the performance against the written reference — with the technical limitations of 2011, which didn’t allow much more than low-difficulty pieces and relatively simple comparisons.
La Nueva España covered the project on 4 January 2011, with the angle of the bagpipe’s chanter (punteru) turned into a controller, acting as a computer pointer — hence the headline. Fifteen years on, the domain gaitahero.com is still alive, and in 2021 the same newspaper name resurfaced in another form: a track on the album Fechu’n PLA that is, at once, a song and a playable video game, conceived as a direct nod to that 2011 headline — I go into more detail about it in the project page Gaita Hero — Video game and musical track, within my projects.
I’m not bringing this up out of nostalgia. I’m bringing it up because in 2011 I was already posing, with the limited tools of the time, a problem I can now name with precision.
The real problem: evaluating a performance is an MIR problem
What Gaita Hero proposed, even in its limited 2011 version, is not a video-game problem. It’s a Music Information Retrieval (MIR) problem: given the audio of a real performance and a symbolic reference — a score, a MIDI file — how do you automatically extract where the performance deviated from the reference?
In its most complete form, that problem can be broken down into at least three axes:
| Axis | What it measures | Why it’s hard |
|---|---|---|
| Timing | Whether each note’s onset arrives before, after, or on time relative to the reference | Human tempo fluctuates; there’s no perfect rhythmic grid to compare against |
| Pitch/tuning | Whether the pitch played matches the expected one | Instruments like the bagpipe don’t always follow 12-note equal temperament |
| Dynamics | Whether the intensity of each note is the one intended | Perceived intensity depends on timbre and register, not just signal amplitude |
Gaita Hero, in 2011, did not cover these three axes with that level of sophistication: with the tools of the time, the system was limited to checking whether the student had played a low-difficulty score correctly, closer to a right/wrong check than to a fine-grained analysis of timing, tuning, and dynamics. But it was already pointing in that direction, and each of those three axes first requires solving, for the input audio, the same underlying task I already described in What is automatic music transcription? (AMT): converting the signal into a symbolic representation — which note, when it starts, how long it lasts, at what intensity — before it can be compared against anything. Without transcription there’s no possible comparison. Evaluating a performance is, at bottom, transcribing and then measuring the distance between two symbolic representations: the one played and the one written.
The Asturian bagpipe adds a difficulty of its own to that problem. It’s a continuous-drone instrument — the drone (roncón) sounds the whole time, overlapping the chanter’s melody — which turns it de facto into a case of polyphony even when the musician is playing a single melodic line. Separating which spectral energy belongs to the drone and which to the melody is no minor nuance: it’s the same harmonic-overlap difficulty that makes polyphonic transcription still, fifteen years after Gaita Hero, an unsolved problem.
Fifteen years aren’t nostalgia, they’re a line of research
Between 2011 and today I haven’t dropped that thread. What in Gaita Hero was a concrete implementation for a final-year project — with the technical limitations of 2011 — has become, over the years, the question that underpins a good part of my current work in music computing and my doctoral research on AMT. The question hasn’t changed: comparing a real performance against a reference and extracting useful information from it. What has changed is how much I know about why it’s hard, and what tools are available today to tackle it.
I keep working on tools that pick up that idea. Right now I’m testing a platform of my own, in real trials within EMTI Fervienza, that conceptually continues the problem I posed in Gaita Hero: comparing an Asturian bagpipe or tin whistle performance against a reference and returning timing, tuning, and dynamics deviations. I won’t go into the technical detail of that platform here — it’s in a testing phase and I don’t have consolidated results to document yet — but what I can affirm with confidence is the framework: it’s an MIR problem applied to instrumental performance evaluation, with fifteen years of continuity behind it, not a rootless new idea.
Why I founded EMTI Fervienza
EMTI Fervienza is the traditional-music and innovation school I founded in 2026, focused on Asturian bagpipe and tin whistle. Its public website is emtifervienza.com. Explaining how teaching works there is not the point of this article — that’s a pedagogical decision that lives on the school’s own website, not in this technical notebook. What does belong on labs.tever.es is the research side: Fervienza is also the setting where I test, with real students and under real conditions, whether the ideas I’ve been developing since Gaita Hero are good for something more than an academic project closed in 2011.
That, to me, is the right reading of the whole thread: a 2011 final-year project wasn’t an isolated experiment that got shelved. It was the first attempt — limited, of its time — at answering an MIR research question I’m still working on fifteen years later, now with my own framework to test it in.
References
The references this article draws on, and where to read further:
- La Nueva España (2011). Coverage of “Gaita Hero”. 4 January 2011.
- Gaita Hero. gaitahero.com.
- EMTI Fervienza. emtifervienza.com.
- Benetos, E., Dixon, S., Duan, Z., & Ewert, S. (2019). Automatic Music Transcription: An Overview. IEEE Signal Processing Magazine, 36(1), 20–30.