MXMinglang Xie
Active projects
AI SaaSMusic AIFull-stack

Song2Notation

AI-powered music transcription for turning audio into sheet music, Jianpu, piano roll, MIDI, and portable notation formats.

Progress board

MVP / quality iteration

Public progress

68%

Updated
2026-07-02

Next.jsSupabasePython workerMusicXMLMIDIJianpu

Current focus

  • • Improving readable notation output and right-hand melody presentation.
  • • Tightening the upload, processing, review, playback, and export loop.
  • • Keeping generated score data available in interoperable formats.

Next steps

  • • Run more transcription-quality checks across representative piano recordings.
  • • Polish history reopening, signed downloads, and export reliability.
  • • Prepare a cleaner public product path after the review loop feels dependable.

The idea

Song2Notation is an AI music-transcription product focused on a deceptively difficult workflow: turning a recorded performance into notation that a musician can inspect, edit, play back, and export.

The intended experience covers audio upload, piano-focused transcription, MusicXML and MIDI generation, traditional staff notation, simplified notation (Jianpu), and piano-roll playback.

Product direction

  • Make the transcription state visible from upload through processing and review.
  • Keep generated notation editable and available in interoperable formats.
  • Support readable and accuracy-oriented presentation modes.
  • Let users return to prior transcriptions and regenerate signed downloads.

Current progress

The current work is centered on notation quality and product reliability. The web app already has the shape of an end-to-end transcription flow: upload, process, review, play back, and export.

The next pass is about making the output more musician-friendly: cleaner score rendering, stronger Jianpu presentation, more dependable format exports, and a history flow that makes prior transcriptions easy to reopen.

Engineering focus

  • Keep the worker output and frontend score view aligned around stable score data.
  • Preserve interoperable outputs such as MIDI and MusicXML rather than trapping users inside one viewer.
  • Make progress states, generated artifacts, and failure cases explicit enough to debug.
  • Evaluate quality with real musical examples, not only happy-path demos.