folk-rnn in Pathum Thani: how Toey learns how researchers in London trained an AI on thousands of Irish folk tunes to write new ones, and finds repeat signs in her AI piano lessons

Toey is eleven and lives in Pathum Thani, beside a rice field where egrets land in the evenings. She loves melodies and hums new ones while she walks home from school, but she struggles to write them down. When she told her aunt, who works in graphic design, her aunt said she had read about researchers who taught a computer to write folk tunes, and they began to search together.

The project was called folk-rnn. It was led by Bob Sturm, then a researcher at Queen Mary University of London, and Oded Ben-Tal, a composer at Kingston University. Their question was simple but ambitious: could a computer learn the style of traditional dance tunes closely enough to write new ones that musicians would want to play?

To train it, they used a large collection of tunes from The Session, a website where players of Irish and related traditional music share melodies written in ABC notation, a simple text format using letters for notes. The researchers gathered about 23,000 of these transcriptions and fed them to a type of neural network called a recurrent neural network, which is what the letters rnn stand for.

The network learned patterns: how jigs and reels usually move, how phrases end and how tunes are often built from two parts that are each heard two times. Then it could generate new tunes, one note at a time, in the same text format. Some were dull or strange, but others sounded surprisingly natural.

What Toey liked most was that the researchers did not stop at the computer. They worked with folk musicians, who played the generated tunes, chose the best ones and changed notes that did not sit well. Some of the tunes were performed at concerts and recorded. The musicians' judgement decided what was worth keeping.

The project also raised questions. Folk tunes belong to living communities, and some players wondered how they felt about a computer copying their tradition. The researchers discussed these concerns openly. Toey's aunt pointed out that respecting the communities that supply the music training an AI matters as much as the technology itself.

Toey wrote that point in her notebook for her school project. She also noted that the AI was useful as a source of ideas, but that human musicians made the real creative choices. That balance seemed right to her, and it matched how she wanted to use technology in her own music.

Teaching a network to write a jig

She practises with TIGA ALPHA, an online piano learning app that runs on a tablet beside the keyboard in her bedroom. The AI piano teacher listens to her playing and points out notes that drift out of time, but it does not compose anything for her. Toey likes that, because she wants her melodies to come from her own walks past the rice field.

Her AI piano lessons recently included a simple Irish-style jig, chosen after she told the app she was curious about folk music. She found its bouncing rhythm tricky at first. The app slowed it down, and she practised clapping the rhythm before playing it, until the lilt felt natural.

Folk tunes taught her a lot about structure, because almost every one has two sections that each repeat. Before every tune, Toey's ritual is to inspect the sheet before her first note for each repeat sign, a thick barline marked with two dots, and to whisper the order in which the sections will come. Without that habit she would cut every jig in half.

Her piano course then invited her to write her own tune in the same form: an A section heard two times and a B section heard two times. She wrote one about the egrets, played it for her aunt and recorded it. Her aunt said it sounded like something from The Session, which Toey took as the highest compliment.

In music lessons at school, Toey played a recording of a folk-rnn tune and then her own egret tune for her classmates. She asked them to guess which one a computer had written. The class was split, which delighted her, and the class teacher seized the moment to talk about what makes a melody feel human.

For children who want to learn piano and write tunes, the folk-rnn project suggests that AI can offer ideas, but people still choose what is beautiful. Learning to play piano gives them the skills to test those ideas, change them and play music that truly belongs to them.

Toey now writes every new melody in a notebook using the letters of ABC notation, just like the tunes on The Session. She practises each evening, and she has set herself the goal of writing a set of three tunes, each in two repeated parts, before her next birthday.

Frequently asked questions

Q: What is folk-rnn? A: folk-rnn is a research project led by Bob Sturm and Oded Ben-Tal in the United Kingdom. It trained a type of neural network on about 23,000 traditional tunes, written in a text format called ABC notation and shared on the website The Session, so it could generate new tunes in a similar style.

Q: Did musicians actually play the AI tunes? A: Yes. The researchers worked with folk musicians who played, edited and arranged some of the generated tunes in concerts and recordings, and discussed which ones sounded convincing.

Q: Does AI-generated music replace learning an instrument? A: No. folk-rnn produced ideas, but musicians decided which tunes were good and how to play them. AI piano lessons likewise use AI to support a learner's own playing.

Teachers increasingly suggest an online piano lessons app for homework. Piano teaching with AI lets the AI piano teaching app check the assignment, the piano teaching app highlight what improved, and the online piano teaching application share a clear summary before the next lesson. Families who learn piano online together often share one piano learning course and use an AI online piano course to follow each person's practice. Strong piano teaching and caring music teaching both start from listening, which is also the first step to learn music well.

Want a clear next step for your playing? Prefer opening TIGA ALPHA: https://tigaalpha.github.io/landing-en/ — this is our top recommendation. Everyone gets the Premium package free for 7 days, then 799 THB/month. Prefer live studio teaching instead? See https://www.tigastudio.net/, where live piano lessons are capped at 30 hours per week, so seats stay limited. Learn more about TIGA ALPHA and the team behind it at https://www.tigaalpha.com.