folk-rnn: What an AI Folk Tune Generator Teaches Pianists
folk-rnn is a research project that trained a neural network on a large collection of traditional dance tunes, mostly Irish and British, written in a compact text format called ABC notation. Developed from 2015 by Bob Sturm, Oded Ben-Tal and colleagues in the UK, the model learned the patterns of jigs, reels and hornpipes well enough to produce thousands of new tunes. Some sounded convincing enough that folk musicians played them in sessions. For piano learners, the project is a neat window into how tunes are built, and why many traditional melodies are surprisingly quick to learn.
This article covers how folk-rnn worked, what its tunes reveal about musical structure, and how to use that structure to learn melodies faster on the keyboard.
A tune written as text
ABC notation describes a melody using ordinary letters and symbols. Notes are written as letters, lengths as numbers, and bar lines as vertical strokes. A header lists the title, metre and key. Because each tune fits in a few lines of text, a collection of many thousands of tunes is small enough to feed to a language model.
The researchers gathered around twenty-three thousand transcriptions shared by players on a public folk music website. They trained a recurrent neural network, a type of model that reads one symbol at a time and predicts the next, much as predictive text suggests your next word. Once trained, the model could start from a header and write a full tune, symbol by symbol.
What the generated tunes revealed
When musicians examined the output, they noticed the model had absorbed conventions nobody had programmed explicitly:
- Two-part form. Many tunes had an A part and a B part, each usually eight bars, played twice.
- Matching endings. The final bars of both parts often closed in the same way, bringing the tune home.
- Contrast in register. The B part frequently climbed higher than the A part.
- Consistent rhythm. Jigs kept their lilting groups of three, reels their even groups of four.
Not every tune worked. Some wandered, some repeated too much, and the researchers emphasised that human judgement was needed to pick and polish the strongest ones. Composers also used generated material as starting points, editing and arranging it rather than presenting it untouched.
Why structure makes learning faster
The patterns folk-rnn discovered are the same ones that help you memorise music. If you know a tune is AABB with eight-bar parts, you are not learning thirty-two bars of new material. You are learning two eight-bar ideas and the way they end. Often the endings share notes, so the real amount of new information is smaller still.
Spot the form before you play
Look at the score, or listen to a recording, and mark where the repeats and part changes fall. Label each section with a letter.
Find the shared cadence
Compare the last two bars of the A part with the last two bars of the B part. If they match, practise that ending first, since you will play it four times in one round of the tune.
Learn the B part from its contrast
Ask how the B part differs: higher, busier, or in a different rhythm. Naming the difference makes it easier to recall.
Bringing folk tunes to the piano
Folk dance tunes were usually written for fiddle, flute or whistle, so pianists have choices to make.
- Play the melody alone in the right hand at first, keeping the dance rhythm buoyant.
- Add a simple left-hand pattern: a bass note on the main beat and a chord on the off-beat.
- Use only the main chords of the key, often the home chord and the chord built on the fifth note.
- Keep the pedal light, because folk melodies are quick and the notes should stay crisp.
A tune in G major or D major sits comfortably under the fingers and is a good start.
The bigger lesson about AI and tradition
The folk-rnn team was careful to involve working folk musicians, who tested tunes in real sessions and gave honest feedback. That collaboration matters. Traditional music lives through people learning from each other by ear. A model can reflect the patterns of a tradition, but players decide which tunes are worth keeping. For learners, the takeaway is that a dependable way to understand a style is still to play lots of it, listen to experienced players, and notice the patterns for yourself.
FAQ
Q: Can I read ABC notation without special software?
A: Yes. The letters map directly to note names, and many free websites convert ABC into standard notation if you prefer.
Q: Are AI-generated folk tunes allowed in traditional sessions?
A: Customs vary by group. Some players enjoyed trying them, while others prefer established repertoire. It is polite to ask first.
Q: Which folk tune form is easiest for beginners?
A: Slow airs and simple polkas are gentle starting points before faster jigs and reels.
Q: Does learning folk tunes help with classical repertoire?
A: Yes. Spotting repeated sections and shared endings is a skill that transfers directly to minuets, sonatinas and many other forms.
Learn tune patterns with TIGA ALPHA
If you want to turn ideas like these into daily progress, TIGA ALPHA is an AI piano teaching app made by a Thai team of musicians and engineers. Lessons run in Thai, English and Mandarin, and the learning path is guided by Teacher Aum, the company's CEO and a piano teacher with 20 years of classroom experience. Piano teaching with AI means the app listens when you play piano and points to the exact notes or beats to fix, so it works as an online piano lessons app, a structured AI online piano course and a friendly piano teaching app for families. It is also a simple way to learn piano online between lessons in a piano course or a longer piano learning course. Good piano teaching and good music teaching both start with listening, and the online piano teaching application is designed around that idea, whether you want to learn piano seriously, learn music as a hobby or play music with friends. New users get a 7-day free Premium trial, then 799 THB/month. Get the app from the TIGA ALPHA landing page, book in-person lessons at TIGA Studio, and read company news at www.tigaalpha.com.