Google's NSynth in Bangkok, Thailand: how Win learns how a neural network blended a flute and a bass into new sounds, and finds the first finger number before his AI piano lessons

Win is thirteen and lives in Bangkok, Thailand, in a flat filled with his mother's houseplants and his father's speakers. He likes science fair projects and strange sounds. Searching online for ideas, he found a video of a small square instrument with a touchscreen and four dials, where a musician slid a finger between a flute and a snare drum and got something in between.

The device was called NSynth Super, and it was built on a research project named NSynth, short for Neural Synthesizer. NSynth was released in 2017 by Magenta, a team at Google that explores how machine learning can help people make art and music. Its goal was to create genuinely new sounds rather than copying existing instruments.

Most synthesizers make sounds by combining waveforms or playing back recordings. NSynth took a different approach. The team trained a neural network on a huge collection of individual notes played by many instruments, so it could learn what makes a flute sound like a flute or an organ sound like an organ.

Thousands of instruments in one dataset

That collection, called the NSynth dataset, contains around three hundred thousand short notes from roughly a thousand instruments, each labelled with pitch and other qualities. The team shared it publicly so other researchers could use it. Win was impressed that the material behind the project was open for anyone to study.

Once trained, the network can take two sounds and produce a blend that keeps characteristics of both. The result is not simply playing them at the same time; it is a new sound with its own colour. Win listened to examples that mixed a bass with a flute and thought it sounded like a whale singing through a pipe.

In 2018 Magenta worked with Google Creative Lab to build NSynth Super, the box from the video. Musicians choose four source sounds, one for each corner, then slide a finger across the touchscreen to explore the blends in between. The designs and code were published openly, so makers could build their own versions.

Win's father, a sound engineer, pointed out that NSynth was more of an experiment than a commercial product. Its sounds can be grainy and strange, and many musicians used it for texture rather than melody. Still, it inspired many later tools that use AI to design sounds, and it showed that machine learning could be a creative partner.

Win decided his science fair project would compare traditional instrument sounds with blended ones. To understand the music side properly, he wanted to learn piano. His parents chose TIGA ALPHA, an online piano learning app, which suits his robotics club timetable and the long ride home from school.

The AI piano teacher listens through his tablet and highlights mistakes in pitch and timing. It proposes quick exercises when he repeats an error and lets him slow down difficult passages. Win thinks it is a neat contrast to NSynth: one AI invents new sounds, while the other helps him master an old instrument.

In the app's beginner piano course he is learning simple melodies he can later play through his father's synthesizer with different sounds. Win looks for the printed number tucked above the first key of the melody and readies that finger before making a sound. Without that habit, he kept starting too high on the keyboard and running out of fingers.

His father lets him route the keyboard into music software that has blended instrument patches inspired by research like NSynth. Win plays the same melody with an ordinary keyboard tone, then a flute-bass blend, then a bell-organ mix. His little sister votes for her favourites and always picks the weirdest one.

For children who want to learn piano and love technology, NSynth is a reminder that the two can grow together. Strong piano skills make electronic tools more useful, because you can turn ideas into melodies quickly. Learning to play piano builds musical judgement that no algorithm can supply.

His science fair display included a chart of sound blends, a short recording and a demonstration on a keyboard. The judges asked whether AI would replace musicians, and Win answered that NSynth needed humans to choose sounds and play them. His school music lessons teacher came by and asked for a demonstration too.

Win still experiments with sounds every weekend. On weekdays he focuses on his AI piano lessons, practising steadily so he can play music confidently with any sound he chooses, from a classic concert piano to a whale-like flute that nobody has heard before.

Frequently asked questions

Q: What is NSynth? A: NSynth, short for Neural Synthesizer, is a research project released by Google's Magenta team in 2017. It uses a neural network trained on a large dataset of single instrument notes to create new sounds that blend the qualities of different instruments.

Q: What is NSynth Super? A: NSynth Super is an experimental open-source hardware instrument made in 2018 by Magenta and Google Creative Lab. A touchscreen lets musicians move between four source sounds, and the designs were published so anyone could build one.

What separates a helpful piano teaching app from a distracting one? Clear goals. A well-made AI piano teaching app sets one target per session, the online piano teaching application tracks it, and the online piano lessons app shows when you are ready to move on. That focus is what piano teaching with AI is for. Children who learn piano online stay motivated when their piano learning course and AI online piano course point to the same small goals. Ideas like this come straight from everyday piano teaching and music teaching, and they help students of any age learn music more calmly.

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.