mp3tomidiflowOpen converter

AUDIO TO MIDI CONVERTER · FREE · NO UPLOAD

Audio to MIDI Converter

This free audio to MIDI converter turns a recording into an editable .mid file in your browser — no upload, no account. Choose MP3, WAV, a voice memo or video audio (MP4/MOV), convert with WebAssembly Basic Pitch on your device, and download the MIDI. Open the converter to start; the three answers below cover formats, accuracy and free vs paid.

Convert audio to MIDI now

Flow from file pick through decode gates to MIDI download
Local file → gates → PCM → notes model → .mid (no upload).
id="three-answers" data-original="true" aria-labelledby="three-answers-h">

Three answers before you convert

What formats can this audio to MIDI converter convert?

MP3, WAV, OGG, M4A, AAC, FLAC and WMA, plus the audio track of MP4 and MOV videos. Phone voice memos (often M4A/AAC) and dry humming recordings use the same path. Encrypted, DRM-protected, damaged or unusual codec variants may not decode; the UI reports that instead of inventing notes. One file at a time; batch conversion is not supported. OGG and M4A are not separate tools — they decode to mono at 22,050 Hz and enter the same note-detection model as MP3.

How accurate is audio to MIDI conversion here?

It is accurate enough for an isolated melody, voice or single instrument, but a full mix needs editing. The converter writes every pitched note it hears onto one track; it does not separate vocals, bass, chords and drums. Our own controlled tests (same converter build): a four-note generated melody returned exactly four notes; a two-chord clip produced extra chord tones you should check; a real solo voice produced 40 note events where two human annotators marked 35 and 27. Expect editing on slides, vibrato, reverb and overlapping instruments. Details and links are in the accuracy table below.

Is this audio to MIDI converter free or paid?

It is free: no account, no payment and no daily quota. Each file can be up to 50 MB and about 3 minutes after decode. Speed depends on your device; keep the tab open until the result or an error appears. There is no paid tier on this page and no credit pack that unlocks the core convert button.

Decode pipeline: exact gates in this build’s app.js

Your file is accepted only if the extension is in a fixed allow-list, size ≤ 50,000,000 bytes, and (after decode) duration ≤ ~180 seconds with non-silent RMS. Those numbers are read from the shipped app.js on this site, not marketing copy.

Method note: This is this page’s own method, not an industry standard.

Hard gates before MIDI is written (source: live app.js on this deployment, 2026-10-26)
GateExact rule in codeUser-visible failure
Extension allow-listmp3 wav ogg m4a aac flac wma mp4 mov (one file)Unsupported format / batch refused
Byte sizefile.size > 5e7 → reject (50,000,000 bytes)Larger than 50 MB
DurationBrowser path: AudioBuffer.duration > 180.01; FFmpeg path: -t 180.1 then length check > 180.01 sLonger than 3 minutes
SilenceRMS of mono 22,050 Hz float samples < 1e-5Silent or too quiet — no MIDI
NotesAfter Basic Pitch, zero valid note events → error notesNo reliable notes — no MIDI invented

Decode order: try the browser’s AudioContext.decodeAudioData first; on failure load bundled FFmpeg WASM, verify the core blob SHA-256 against ffmpeg-core.json, then run -i input.EXT -t 180.1 -vn -ac 1 -ar 22050 -f f32le out.raw. The model always sees mono float32 at 22,050 Hz. MIDI is written locally with division = 480 PPQ and velocity = round(amplitude × 127) clamped to 1–127.

What you can verify here: the exact size, duration, silence, and “no notes → no MIDI file” reject rules above; the FFmpeg argv used when the browser decode path fails; and the asset list in No-upload proof you can re-check in DevTools. This page publishes those numbers for this build. It does not invent notes when detection returns zero events.

Voice to MIDI on this same converter

Yes — a single dry voice or hum converts on this page; a voice inside a full song does not get pulled out automatically. Record a phone voice memo (M4A, MP3, WAV or AAC under 50 MB / 3 minutes), open the converter, choose the file and download the .mid. Humming works when notes are clear and separated; long connected hums blur note boundaries. The MIDI stores fixed semitone pitches with no pitch-bend, so scoops, slides and wide vibrato can become a short extra note, a split note or a neighbouring semitone.

What works: monophonic singing or humming with no backing track, neutral syllables (“da”, “doo”, “la”), quiet room. What fails or needs another tool first: extracting a lead vocal from a finished mix (no stem split here), choir/harmony stacks, talk-over with music. For a dedicated vocal walkthrough keep vocal to MIDI; do not look for a separate /voice-to-midi URL — that intent is covered here and on the vocal page.

Voice to MIDI: when this converter is the right tool
InputUse this page?What you get
Phone voice memo, dry soloYesOne track of detected pitches to edit in a DAW
Hummed melody, clear note startsYesUsually cleaner than legato singing
Solo vocal with light room reverbYes, then editExtra short notes around sustained tones are common
Lead vocal inside a full mixNo (not alone)All pitched parts land on one track; isolate the vocal elsewhere first
Choir / stacked harmoniesPoor fitOverlapping pitches become a dense single track

Accuracy: what our own tests returned

None of these numbers are “100% accurate” claims — they are measured outputs from this site’s converter on named inputs. Use them to set expectations before you convert a song.

Controlled tests on this converter (not a rival benchmark)
InputWhat we measuredWhere the detail lives
Four-note generated melody (WAV and MP3 of the same line)Exactly four notes returned for that melodyWAV vs MP3 same-melody test
Two-chord test clipExtra chord tones appeared — check and pruneExtra chord notes
Real solo voice recording40 note events; two human annotators marked 35 and 27Real vocal example

Practical rule: start with an isolated part or stem. A full song to MIDI is allowed within the size/length limits, but the result is a dense single track you will edit. If you need separate bass, chord and melody tracks from a mix, a stem-splitting tool is the better first step (see the practical comparison below).

What this audio to MIDI converter does

You give it one audio file. It detects pitched notes (pitch, start time, length and a velocity from loudness) and writes them into a standard single-track MIDI file. The MIDI stores instructions, not the recorded sound, so playback later depends on the instrument you assign in a DAW or notation editor.

The same converter handles every supported format on one page. Format-specific tips live on WAV to MIDI, vocal to MIDI, piano to MIDI, guitar to MIDI and the step guide how to convert audio to MIDI. For both audio→MIDI and MIDI→audio directions, see MIDI converter.

No-upload proof: what loads, and what never leaves

Competitors often claim privacy; few show you how to verify it. Here is what this site's converter actually fetches from our own origin when you convert, based on the live app.js and the files under /vendor/ on this deployment.

What the browser loads from this site

Assets fetched from mp3tomidiflow.com during conversion (not your audio)
PathRoleApprox. size on this build
/app.jsConverter UI and orchestration9,048 bytes
/vendor/transcribe.jsWeb Worker that runs note detection1,176,739 bytes
/vendor/model/model.json + group1-shard1of1.binBasic Pitch model weightsmodel shard 742,392 bytes
/vendor/wasm/*.wasmTensorFlow.js WASM backends (incl. SIMD variants)e.g. tfjs-backend-wasm.wasm 275,675 bytes
/vendor/core/ffmpeg-core.js + ffmpeg-core.part0.bin + ffmpeg-core.part1.binBundled FFmpeg WASM decoder (used when the browser cannot decode the file itself)parts total 32,232,419 bytes (sha256 in ffmpeg-core.json)

These requests are GETs for tools this site hosts. They are not your recording. The model is Spotify's open-source Basic Pitch, packaged to run locally via WebAssembly.

What never leaves the device

How to check in DevTools that no audio request is sent

  1. Open the converter in Chrome, Edge, Firefox or Safari.
  2. Open DevTools → Network. Enable Preserve log. Clear the list.
  3. Choose a short test file (for example a few seconds of WAV or MP3 under the 50 MB / 3 minute limits) and click Convert to MIDI.
  4. Watch the Network list while status text moves through “Reading your audio…”, “Decoding audio…”, “Loading the note-detection model…” and “Detecting notes…”.
  5. Expected: requests whose URLs stay on mp3tomidiflow.com (or the same origin) for paths under /app.js, /vendor/transcribe.js, /vendor/model/, /vendor/wasm/ and /vendor/core/. Method is GET. Payloads are the tool files listed above.
  6. Not expected: any request whose request body or FormData contains your audio filename or PCM/MP3 bytes, or a POST/PUT of the file to another host. Filter by your filename in the Network search box — it should not match a network transfer of that file.
  7. When finished, the result line shows something like N notes · X.X seconds and a Download .mid link. The .mid comes from a blob: URL, not from a server response that uploaded your audio.

Site statistics (Google Analytics 4, see the privacy policy) are separate from conversion and never receive your audio file.

We observed this flow against the shipped app.js: size gate at 50,000,000 bytes; duration gate after decode at about 180 seconds (-t 180.1 in the FFmpeg args, then a length check); silent-input rejection when RMS is below a fixed threshold; and error strings that refuse to invent MIDI when no notes are detected.

How this converter differs in practice

Input: one audio file you convert here (browser, no upload) versus what other public converters state on their own pages about upload, account, stems, and limits.

Result on this page: a single-track .mid from detected pitches; published file list + DevTools check that no audio POST leaves the device; voice-to-MIDI decision table; controlled accuracy tests on named inputs; hard gates at 50 MB / about 3 minutes / non-silent RMS / zero-notes refuse.

Basis: what this site’s shipped app.js and /vendor/ assets actually do, plus what each comparison row below quotes from that site’s own page text. “Not stated” means we did not find it on that page — not a claim the feature is absent from their product.

When this converter is not the right tool: you need automatic stem split (bass / chords / melody / drums as separate tracks), a multi-instrument mix cleaned into parts, or a paid cloud model with account credits. Several other tools advertise those paths; this page does not.

Song to MIDI and music to MIDI: what to expect

You can convert a song to MIDI here, but a full mix is the hardest input — see how accurate conversion is and the accuracy table. For a dry voice or hum, use the voice to MIDI section on this page. For music to MIDI, start with an isolated part or a stem if you have one, and expect to edit. If you want notation rather than a MIDI file, see audio to sheet music.

Limits

Fix common problems

Match the symptom to one check, then take the next step below — this page has no sensitivity knobs and no invented accuracy %. Limits stay 50 MB and about 3 minutes; note detection is local Basic Pitch–style pitch tracking on one track.

Symptom → check → next step (this site’s real gates)
What you seeCheck firstNext step
No notes detected / no downloadFull mix, drums, heavy reverb, damaged or DRM-protected fileUse a clean single melody; drums → drum to MIDI; dry voice → vocal to MIDI
Notes appear but many are missingSeveral instruments overlappingTrim a 15-second trial first; isolate the part, then convert again
Whole line is an octave off—Transpose the MIDI up or down by an octave in your DAW, then edit
File refused / will not convertOver 50 MB or over 3 minutes, or unsupported formatShorten or compress; for MP3→WAV prep use MP3 to WAV
WAV drifts in video editors / wrong sample rateTimeline needs 48 kHzMP3 to WAV Video preset (48 kHz · 16-bit · stereo)
Speech transcription sounds poor after exportStill on 44.1 kHz stereoMP3 to WAV Speech preset (16 kHz · 16-bit · mono)

Local pitch is the main pitch track, not stem separation. The in-browser Basic Pitch–style model writes every pitched note it hears onto one MIDI track. It does not split vocals, bass, chords or drums into stems. When the model returns zero usable note events, the UI refuses to invent MIDI (“No reliable notes were detected… No MIDI was generated”) and does not offer a download — that refuse path is intentional, not a broken button.

15-second trial workflow: (1) Cut a clear 15-second slice where one melody or instrument is audible. (2) Open the converter, convert that slice, and listen to the .mid in a DAW. (3) If the notes look usable, convert a longer take of the same isolated part. (4) If the slice fails or is dense garbage, fix the source (isolate, dry room, shorter) before spending time on the full file. Pick a starting path on the audio tools hub when you are unsure which tool fits.

Real test note: four-note generated melody → Download .mid offered with exactly four notes; intentional no-pitch / silent or unpitched cases that hit the notes gate → “No reliable notes were detected… No MIDI was generated,” download not offered (same refuse path documented on MP3 to MIDI not working). Dense full mixes usually still offer a download of a crowded single track you must edit — use isolation or drum to MIDI / vocal to MIDI / piano to MIDI / guitar to MIDI instead of expecting a clean full-song transcription.

How to convert, step by step

  1. Open the converter. No install, no account.
  2. Choose one audio or video-audio file within the limits above.
  3. Convert and keep the tab open. Optionally open DevTools Network first if you want to verify the no-upload behaviour.
  4. Download the .mid and edit it in a DAW. See what is in your MIDI download and how to convert audio to MIDI for checking notes before you edit.

Questions

Is this audio to MIDI converter free?

It is free: no account, no payment and no daily quota. Each file can be up to 50 MB and 3 minutes long. There is no paid unlock for the core convert button on this page.

Is my audio uploaded when I convert audio to MIDI?

No. Decoding and note detection run in your browser with WebAssembly Basic Pitch. The site only delivers the converter, model and decoder files. You can confirm in DevTools Network that no request sends your audio file.

Which formats does the audio to MIDI converter accept?

MP3, WAV, OGG, M4A, AAC, FLAC and WMA, plus the audio track of MP4 and MOV videos. Encrypted, DRM-protected, damaged or unusual codec variants may not decode.

How accurate is this audio to MIDI converter?

It is accurate enough for an isolated melody, voice or single instrument, but a full mix needs editing. Our tests: four-note melody → exactly four notes; two-chord clip → extra chord tones; real solo voice → 40 note events vs human marks of 35 and 27. See accuracy tests.

Can I convert voice or humming to MIDI?

Yes for a single dry voice or hum under 50 MB and 3 minutes. A voice inside a full song is not separated automatically. See voice to MIDI and vocal to MIDI.

How do I verify that no audio is uploaded?

Open DevTools Network before you convert. During conversion you should see requests only for this site's script, model and decoder files under paths such as /app.js, /vendor/transcribe.js, /vendor/model/ and /vendor/core/. Your chosen audio file should not appear as a POST or PUT body to any server.

Can I convert a whole song to MIDI?

Yes, but expect editing. The converter puts every detected note from the mix on one track and does not separate instruments, so a full song gives a dense MIDI. An isolated melody, voice or instrument part gives a much cleaner result.

Format → decode gate → MIDI readiness (this build)

Input: the container/codec you pick (WAV, MP3, M4A/AAC, OGG/Opus, FLAC, or an oversize / unsupported file).

Result: either a PCM analysis buffer ready for pitch/onset, or a blocked gate with no MIDI started. Rows below mirror checks in this site’s browser pipeline (see also the decode pipeline).

Basis: decode behavior and size/duration gates in this converter build — measurable in the page before notes are written.

When not applicable: a green decode gate is not a promise of clean notation. Soft onsets, polyphony, and reverb can still produce messy note lists; use the symptom matrix in Fix common problems.

These limits and checks are how THIS converter behaves — not a claim about every DAW.

Container/codec gate → what this build does next
Input you pickBrowser decode gateIf gate passesMIDI pitch/onset readinessHonest limit
WAV (PCM / float)Web Audio decodeAudioData on the file bytesMono-mixed analysis buffer in memoryReady for onset + pitch pass on the trial windowHuge WAVs still hit the size/duration ceiling before decode
MP3Same decodeAudioData path after container sniffDecoded PCM frame bufferReady if the MP3 is not truncated/corruptSome rare MP3 packs fail decode in one browser and pass in another
M4A / AACdecodeAudioData when the browser ships an AAC decoderPCM buffer like WAV/MP3Ready on Chromium/Safari paths that expose AACFirefox builds without AAC support refuse the file at the gate
OGG / OpusdecodeAudioData when Opus/Vorbis is availablePCM bufferReady on supporting browsersUnsupported browser → refuse before any MIDI notes
FLACdecodeAudioData where FLAC is built-inPCM bufferReadyOlder Safari/Chrome gaps: convert via MP3 to WAV / WAV export first
Oversize / overtime fileSize > 50 MB or duration > 3 minutes checkBlocked — no decodeNot startedTrim a 15-second trial; see fix common problems
Unsupported extension / empty fileExtension + empty-byte guardBlockedNot startedRe-export as WAV/MP3; hub index at audio tools
Stereo mix with heavy bleedDecode still passesBuffer builtPitch pass runs but note list is noisyGate ≠ musical quality — isolate a dry melody first

How to read the table: decode gate is binary (file becomes PCM or it does not). MIDI readiness only means the pitch/onset pass is allowed to run — it is not a promise of clean notation. Soft onsets, polyphony, and reverb still fail after a green gate; use the symptom matrix below for those cases.

Related same-topic pages that feed this converter: WAV to MIDI, vocal to MIDI, piano to MIDI, guitar to MIDI, MIDI converter hub.

Error category → on-page message (this build)

Input: a conversion failure thrown inside this page’s browser pipeline.

Result: one of five labeled categories (length, silent, notes, model, decode) mapped to a fixed user-facing sentence — or a generic fallback.

Basis: the failure map in this site’s app.js convert path (same categories used for analytics tags). Measurable by forcing each gate with a test file.

When not applicable: a green result panel means notes were written; this table does not describe DAW import quirks after download.

These messages are how THIS converter behaves — not a claim about every audio-to-MIDI product.

app.js failure category → user message (2026-10-29 deepen)
CategoryWhat tripped it in this buildUser-facing message (verbatim intent)What to try next
lengthDecoded duration > 180.01 sRecording longer than 3 minutes — choose a shorter clip; nothing convertedTrim a 15–60 s melody trial
silentRMS of analysis buffer < 1e-5Silent or too quiet — no MIDI generatedRaise input level; avoid empty takes
notesModel returned zero usable note eventsNo reliable notes detected — try a clear melody or single instrumentIsolate a dry lead; see accuracy tests
modelNote-detection worker error / unloadNote-detection model could not finish — try shorter clip or reloadReload; disable aggressive script blockers for this origin
decodeffmpeg-core / decodeAudioData / corrupt buffer path failedAudio could not be decoded — damaged, encrypted, or unsupported codecRe-export WAV/MP3; check format decode gate
other / genericUnlabeled throwConversion could not finish — try a shorter, unprotected fileRetry after reload; confirm file < 50 MB

How to read the table: categories are mutually exclusive labels on a failed run. A successful download never emits these strings. Related: fix common problems, decode pipeline, audio tools hub.

Onset / pitch-confidence triage (this build)

Input: a finished or failed conversion status from this browser tool.

Result: which triage lane to try next — without uploading audio anywhere.

Basis: the same on-page gates as the error-category map and format-decode sections (length / silent / notes / model / decode).

When not applicable: after a clean MIDI download, use the export checklist below instead of this triage.

Audio to MIDI confidence triage flow
Decode → onset density → monophonic vs dense mix expectations.
Status → user fix (2026-10-30 deepen)
Observed statusLikely cause on this pageWhat to try
Decode / unsupported codec messageBrowser path could not produce PCMRe-export WAV/MP3; see format decode gate
Silent / too quietRMS under analysis floorRaise gain; avoid empty takes
No reliable notesModel returned zero usable eventsDry single-melody clip 15–60 s
Sparse MIDI, melody mostly thereOnset density low — monophonic-friendlyExport .mid; light quantize in DAW if needed
Dense mix → many missing / wrong pitchesPolyphony beyond what this build claimsIsolate lead; do not expect full-band transcription

Polyphony vs monophonic expectation matrix

What this browser converter can reasonably claim — not a studio stem separator.

Material type → expectation on this tool
MaterialExpectationNot a claim
Solo voice / whistle / single synth leadOften usable MIDI outlinePerfect pitch every note
Solo piano (clear, dry)Partial melody + some harmonyFull two-hand score
Guitar single-note lineOften usable if dryChord voicings complete
Full band / dense EDMExpect misses & octave errorsMulti-track separation
Drums-only groovePrefer drum-to-midi pathMelodic pitch track

Export-target checklist (after .mid downloads)

  1. Confirm the file downloaded locally (no upload step on this page).
  2. Open in a MIDI player or DAW; check note count vs what you heard.
  3. Verify tempo / PPQ if the DAW asks — do not assume 120 BPM matches the source.
  4. Quantize only if needed; over-quantize hides timing that was actually detected.
  5. If the lead is wrong but rhythm is right, re-run with a drier, shorter clip rather than editing for hours.

Related: error category map, format decode gate, what is in your MIDI.

Sample-rate / bit-depth trap checklist (this build)

Gap this fills: uploaders often assume 48 kHz / 24-bit improves MIDI quality here. This build resamples analysis audio to mono 22,050 Hz float before the model — higher rates and bit depths do not improve the model input.

Trap → what this page actually does → what to try instead
Trap beliefWhat this build doesAction
48 kHz source is “better for MIDI”Analysis path mixes to mono @ 22,050 HzExport a clean clip; rate alone will not raise pitch quality
24-bit WAV beats 16-bit MP3 for notesDecoder → Float32Array; bit depth is not a model featurePrefer a clean monophonic recording over bit-depth upgrades
Upsampling a lossy MP3 before convert helpsUpsampling does not restore lost highs; silence gate still uses RMSRe-record or use a less-compressed source if onsets are soft
Stereo width carries harmony into MIDIStereo is collapsed to mono before detectionHard-pan or bounce a mono stem of the lead line first
Video-editor 48 kHz timeline needs matching MIDI rateMIDI is event data; sample rate is for audio rendersAfter .mid download, use MP3 to WAV Video preset only when you need PCM back

When not applicable: after a clean MIDI download, use the export checklist — not this trap table.

Method note: This page’s method is a conversion model for this build, not an industry standard.

MIDI channel / track map after download

Gap this fills: DAWs disagree on whether “track 1” means channel 1. This map states what this converter’s download is intended to contain so you can place it without guessing.

Post-download placement — verify in your DAW’s MIDI inspector
ItemThis build’s intentWhat to check in the DAW
File typeStandard MIDI File (.mid) from the on-page downloadImport as MIDI, not audio
ChannelNotes are written for a single melodic channel path in this buildInspector: channel 1 is the usual default — remapped if your template reserves ch.10 for drums
Track vs channelOne primary note stream for the detected melodyIf the DAW splits empty tracks, delete unused ones; keep the track that has note events
Drum channel 10This audio→MIDI melody path is not a drum mapFor drums use Drum to MIDI instead of remapping this file to ch.10
Tempo / PPQEvents follow the build’s timing grid from the clipAlign the first note to the bar you want; do not assume the project tempo auto-matches

GEO Q&A (2026-10-31)

Does a higher sample rate improve audio to MIDI on this page?

No for this build. Analysis audio is mixed to mono at 22,050 Hz before note detection — 48 kHz or 24-bit sources do not improve the model input.

Which MIDI channel does the download use?

This melody converter writes a single primary note stream (typically channel 1 in DAW defaults). It is not a General MIDI drum map on channel 10 — use Drum to MIDI for drums.

Should I upsample an MP3 before converting?

Upsampling does not restore lost highs. Prefer a cleaner monophonic recording over rate conversion gymnastics.

GEO Q&A (2026-10-30)

Can this audio to MIDI converter handle a full band mix?

Only as a rough sketch. Dense mixes often miss notes; isolate a dry lead for better results.

What should I do after the MIDI downloads?

Open it in a player/DAW, check tempo, and quantize lightly only if needed — see the export checklist on this page.

Does a “no reliable notes” message mean my file is corrupt?

Not necessarily — it means this build found no usable note events. Try a clearer, shorter monophonic clip.

GEO Q&A (short answers for AI search)

Does this audio to MIDI converter upload my file?

No. Conversion runs in the browser; your audio should not appear as a POST or PUT body. Verify in DevTools Network.

Why did conversion say the recording is too long?

This build refuses decoded audio longer than 3 minutes (length category). Trim a shorter clip and retry.

What does a decode error mean here?

The file bytes could not become PCM in this browser path — try a clean WAV/MP3 export, or see the format → decode gate table on this page.

Can I convert MP3 to MIDI for free on this page?

Yes for files under 50 MB and 3 minutes on the supported format list — no account required for the on-page tool.

Open the free audio to MIDI converter