Back to blog

Audio to Notation: How to Turn Recordings into Sheet Music

1 November 2026
Audio to Notation: How to Turn Recordings into Sheet Music

Audio to notation software listens to a recording and writes out the notes as a score. You upload an MP3, WAV or video link. The software detects pitches and timing. You get a draft score that you export as MusicXML, MIDI or PDF. The draft is rarely finished. Plan to check it by ear and correct it in a notation editor.

This guide explains how the process works, what changes the result, how to choose a tool, and how to turn a rough draft into a score you can hand to a student or an ensemble.

TL;DR

  • Audio to notation tools detect pitch and timing, then build a score. The output is a draft, not a finished edition.
  • Clean, solo recordings give the best results. Dense, multi-instrument mixes give the worst.
  • Export to MusicXML if you want to edit the score. Export to MIDI if you only need the notes and timing.
  • Check every draft against the recording. Look at pitches, rhythms, rests, bar lines and time signatures.
  • If the draft is unusable, transcribe by ear. A poor draft can still save time as a starting point.

Table of Contents

How audio to notation works

Most tools follow the same sequence. The names differ, but the stages are similar.

  1. Input. You upload a file or paste a link. Some tools accept a live recording from a microphone.
  2. Separation. Some tools split the recording into parts, such as vocals, bass, drums and other instruments. This step is optional and is used for mixed recordings.
  3. Note detection. The software finds pitches and the moments each note starts and stops. The result is often MIDI.
  4. Rhythm and meter. The software guesses the tempo, the time signature and where the bar lines fall. It rounds the timing to note values such as quavers and semiquavers. This is called quantisation.
  5. Engraving. The software lays out the notes as a score and exports it as PDF or MusicXML.

An open-source project on GitHub shows this pipeline in a plain form. Its documentation lists ingestion, optional instrument separation, MIDI transcription, arrangement, humanising and engraving to PDF and MusicXML. Commercial tools hide the stages behind one button, but the same work is being done.

Step 4 is where many errors appear. Detecting a pitch is one task. Deciding that a note is a dotted crotchet and not a crotchet tied to a quaver is another.

How accurate is automatic transcription?

Automatic transcription is useful, but it is not equal to a trained human transcriber.

A review of the field by researchers at Queen Mary University of London states that accuracy remains significantly below that of human experts. It adds that reported accuracy seems to have reached a limit. It also notes that general-purpose models struggle with the variety of music. The authors suggest that systems built for specific uses, and semi-automatic approaches, may help.

A 2021 study in Digital Scholarship in the Humanities compared algorithmic and human transcribers. It found that algorithms make errors that people do not. Examples include reading ornamentation as melody notes, letting accompaniment spill into a melody, and adding notes that are not in the music.

Three common errors in algorithmic transcription

The difficulty grows with the number of instruments. A 2025 multi-instrument transcription challenge compared eight teams. The top model reached an F1 score of 0.5998. That figure shows how hard polyphonic, multi-instrument recordings remain.

New datasets are being built to improve this. One is the Beethoven Symphony Excerpt Dataset, released in 2026. It offers manually verified, time-aligned orchestral data for testing. Orchestral transcription has lacked this kind of benchmark.

What this means for a teacher

Treat the output as a first draft. That is a practical way to work, not a limitation to worry about. A draft gives you the notes in the right order and a layout to edit. Correcting a draft is often faster than writing every note from nothing.

What affects the result

You control several of the factors that decide how good a draft is.

FactorBetter resultWorse result
Number of instrumentsOne instrument or one melodic lineFull band or choir with many parts
Sound qualityClean, close recording with little noiseRoom noise, distortion, heavy compression
EffectsDry soundReverb, delay, chorus
PerformanceSteady tempo, accurate pitchRubato, loose rhythm, out-of-tune notes
Instrument typePiano, guitar or voice with a tool built for itAn instrument the tool does not handle
StyleClear melody and simple rhythmOrnaments, fast runs, swung or syncopated rhythms

A clean source matters. The performer's rhythmic accuracy matters too. If the player drifts in tempo, the software cannot tell a deliberate slowing from a mistake. It will often place notes in the wrong bar.

Research also suggests that a model trained on one instrument can struggle on another. A 2025 paper on transcription between piano and guitar reports that joint training and data oversampling can reduce these gaps. For you, this means a tool built for piano is the better choice for piano, and the same applies to guitar.

Tips for better input

  • Record one part at a time when you can.
  • Place the microphone close to the instrument.
  • Use a metronome or click track so the tempo stays steady.
  • Avoid reverb and effects.
  • Trim the file to the section you need. A 30-second excerpt is easier to check than a full song.
  • Export in a lossless format such as WAV when you can.

Choosing audio to sheet music software

There are many tools. They fall into a few types. Choose by type first, then compare individual products on the criteria below.

Types of tool

  • Instrument-specific apps. These are built for one instrument, such as piano or guitar. They usually handle that instrument best.
  • Multi-instrument transcription services. These accept full mixes and try to separate and transcribe several parts. They are the most ambitious and the least reliable on dense music.
  • Melody extractors. These suit a single line, such as a voice or a solo wind instrument.
  • Desktop transcription programs. These run on your computer and often include an editor.
  • Notation editors that accept MIDI. These do not listen to audio. They help you clean up what another tool produced.

Criteria that matter

QuestionWhy it matters
Which instruments does it handle?A piano tool and a guitar tool give different results on the same recording.
Does it export MusicXML?MusicXML opens in most notation editors and keeps notes, rhythms and layout.
Does it include an editor?A built-in editor lets you fix errors without another program.
Does it handle several instruments, or one line?Many tools work best on a solo line.
What are the file limits?Check maximum length, file types and whether YouTube links are accepted.
What does it cost?Free tiers often limit length or export formats. Check the provider's own page for the current plan.

Professional advice is consistent here. Export to MusicXML or MIDI and refine the result in notation software. Do not expect a print-ready score straight from the tool.

A general chat assistant is not a transcription tool. It works with text, so it cannot listen to an audio file and return a score. It can explain notation and help you plan an edit, but you need transcription software for the audio itself.

From audio to MusicXML to a finished score

MusicXML is the common format for moving a score between programs. If your goal is an editable score, follow these steps.

  1. Prepare the audio. Trim it, remove silence and choose a clean section.
  2. Upload and select the instrument. If the tool asks for the instrument or the type of output, choose it. A wrong setting causes wrong notes.
  3. Export MusicXML. Also export MIDI if the tool offers it. MIDI is useful as a backup.
  4. Open the file in a notation editor. Free and paid editors both read MusicXML.
  5. Fix the structure first. Set the correct time signature, key signature and tempo. Check that bar lines fall in the right places.
  6. Fix the notes. Play the score back beside the recording. Correct wrong pitches, missing notes and extra notes.
  7. Fix the rhythm. Replace awkward note values and tied notes with clear ones. Remove tiny rests and grace notes created by loose timing.
  8. Add the details. Add dynamics, articulation, lyrics, chord symbols and rehearsal marks.
  9. Export the final version. Save a PDF for printing and keep the editable file.

Work in that order. Wrong bar lines make every later check harder. Fix them before you look at single notes.

If you only have MIDI, import it into the editor and clean up quantisation and notation errors. The process is the same, but you may need to set the time signature yourself.

How to check a draft

A draft that looks neat can still be wrong. Checking takes a few minutes if you do it in a fixed routine.

  • Pitch. Play each phrase on an instrument and compare it with the recording. Check the first and last note of every phrase.
  • Rhythm. Tap the pulse while the recording plays. Check that the notes land on the beats shown in the score.
  • Rests. Software often removes short rests or adds extra ones.
  • Bar lines and time signature. Count the beats in each bar. A bar that is too long or too short shows an error.
  • Key signature. Wrong key signatures lead to many accidentals. Choose the key by ear and re-spell the notes.
  • Extra notes. Remove notes that are not in the recording. Research has found that transcribers add non-existent notes.
  • Chords. Check chord tones against the recording, particularly in the bass.

A backing track can speed this up. A plain guide to what a performance track is explains how a prepared track lets a performer play along with a fixed reference. The same idea works for checking a score: play your corrected notation over the recording and listen for any clash.

Checking in a classroom

Checking can be a learning task. Students who find the errors in a draft practise listening, counting and reading at the same time.

Stormusi Classroom has tools that fit this work. Rhythm Player supports rhythm practice, so students can compare a written pattern with what they hear. Pitch Matcher supports pitch work, and Echo Trainer supports listening and repeating phrases. You can set a short passage from a draft as an exercise and ask students to mark what is wrong.

When automatic transcription does not work

Sometimes the output is not usable. The recording may be too dense, the performance too loose, or the instrument unsupported. You have three options.

  1. Transcribe by ear. Work in short phrases. Slow the recording down, loop each phrase and write it in a notation editor. Start with the bass line and the melody, then add inner parts.
  2. Use the draft as a scaffold. The 2021 study found that even low-quality automatic output helped less-skilled transcribers produce better-rated scores than starting from nothing. A rough draft gives you something to fix, and fixing is often easier than writing from a blank page.
  3. Hire a professional transcriber. This suits scores for publication, exams or performance with a fee.

Try a different approach to the same recording as well. Isolate one part, shorten the file, or record a clean version of the part yourself.

Our perspective

Music teachers rarely need a perfect transcription of a full mix. They need a usable score of one part, for one class or one student, in time for the next lesson.

For that reason we think of audio to notation as the first step in a workflow. The tool produces a draft. The teacher checks it. The class uses it. The checking is where students learn the most.

Stormusi is built by educators for educators. Our Classroom tools support the work around a score: rhythm, pitch, listening and repertoire. Our Composer app supports sheet music creation with AI. Our products share one login, and we do not charge per student. You pay only for the products you use.

We suggest that you keep the process simple. Pick a clean recording. Export MusicXML. Edit the draft. Then use it in a lesson and let students find the mistakes.

Stormusi

Build the score, then teach with it

Once the score is correct, the next job is using it. Stormusi supports that part of the work.

Stormusi Composer supports sheet music creation with AI, so you can build and adapt notation for your students. Repertoire Tracker helps you keep track of the pieces your students learn, and Weekly Planner helps you plan lessons around them.

Stormusi

Each Stormusi product has a free tier option. Paid plans are billed per product, with no per-student fee, and bundles carry a discount. You can start from the Stormusi home page.

FAQ

How do I convert audio to music notation?

Upload your recording to a transcription tool and choose the instrument if asked. Export the result as MusicXML or MIDI. Open it in a notation editor, then correct the time signature, notes and rhythms by ear.

Can ChatGPT convert audio to sheet music?

No. A general chat assistant works with text and cannot turn an audio file into a score. It can explain notation or help you plan edits. For the audio itself, use transcription software.

How can I turn audio into notes if the software gets it wrong?

Use the draft as a starting point, or transcribe by ear in short, looped phrases. Research from 2021 found that even weak automatic output helped less-skilled transcribers. Start with the melody and bass. For publication-level work, consider a professional transcriber.

What is the best format to export?

MusicXML is the best choice for editing, because most notation editors open it and it keeps rhythms and layout. MIDI keeps the notes and timing but not the engraving. PDF is for printing and cannot be edited easily.

Does it work for full bands and orchestras?

Not reliably. In a 2025 challenge on multi-instrument transcription, the top model scored an F1 of 0.5998. Work with one part at a time where you can. Solo lines give far better results.

Sources

Recommended