Explainer
What Is Transkun? The AI Model Behind Pianolyze
Jul 20, 2026 · 5 min read
You may have seen Transkun named as the model behind a piano transcription tool or found its GitHub repository. It is the open-source model that powers Pianolyze's transcription.
What Transkun is
Transkun is an open-source transformer model for piano transcription. It converts a piano recording into structured note data: which keys were pressed, when, and for how long. Built as a research project, it focuses on solo piano audio and produces MIDI-equivalent note events.
Transkun is a model and command-line tool. It does not provide Pianolyze’s upload flow, piano roll, or playback controls.
Pianolyze uses Transkun in a server-side transcription pipeline and presents the results in a piano roll with MIDI export.
How it works, briefly
Transkun uses a transformer architecture trained on paired audio-and-note datasets such as MAESTRO. That dataset contains piano performances recorded on a Disklavier, which captures exact note timing alongside the audio. Given a new recording, the model predicts the sequence of notes most likely to have produced it.
The result is strongest on solo or near-solo piano. Like every transcription model, accuracy drops on fast, heavily pedaled passages, and drops sharply on recordings where piano is one instrument among several.
Running Transkun yourself
If you're a developer or researcher, running Transkun directly gives you full control:
- Set up a Python environment and install the dependencies
- Download the model checkpoint
- Run the CLI against your audio files
- Get raw note-event output you can script into a larger pipeline
Run Transkun directly if you are batch-processing an archive, doing research, or integrating transcription into another tool. If you only want to see the notes in a song, the setup and nonvisual output add unnecessary work.
Running Transkun through Pianolyze
Pianolyze uses Transkun on the server as part of a browser-based workflow:
| Raw Transkun | Pianolyze | |
|---|---|---|
| Setup | Python environment, checkpoint download | Open a browser tab |
| Interface | Command line, raw output | Piano roll, instrument isolation, MIDI export |
| Runs where | Your machine, via Python | Server-side analysis |
| Audio upload | None when run locally | Uploaded for analysis |
| Cost | Free (open source) | Free preview; Song Pass or monthly plan |
| Best for | Developers, batch jobs, research | Learning, practicing, one-off transcriptions |
The better option depends on the task. Use Transkun directly to script batch jobs over hundreds of recordings. Use Pianolyze to transcribe a single recording and view the notes without working in a terminal.
For other options, see our comparisons of AnthemScore alternatives and the best AI music transcription tools.
Try Transkun without installing anything
Pianolyze uses Transkun in its server-side transcription pipeline. Preview your own recording for free, then choose a Song Pass or monthly plan.
Open PianolyzeFrequently asked questions
- Is Transkun free?
- Yes. You can run Transkun yourself with Python. Pianolyze uses it on the server and offers a free preview of your own recording after email sign-in; full recordings use a Song Pass or monthly plan.
- What's the difference between Transkun and Pianolyze?
- Transkun converts piano audio into note data. Pianolyze uses it in a server-side pipeline and adds instrument separation, a piano roll, MIDI export, and playback controls.
- Do I need a GPU to run Transkun?
- A GPU speeds up local Transkun runs, but a CPU can also run the model more slowly. Pianolyze processes uploaded audio on its servers, so your computer does not run the transcription model.
- How accurate is Transkun?
- For solo piano, transformer-based models like Transkun benchmark well on standard datasets such as MAESTRO, and the output is generally usable for learning or as a notation first draft. Accuracy drops on fast, heavily pedaled passages and on recordings where piano isn't the dominant instrument.