If the part exists as MIDI, transpose it — that is lossless at any interval. If it exists only as audio, pitch shift with formant preservation on, and expect roughly two to three clean semitones on vocals and four to five on most instruments. Beyond that, the character of the sound changes in a way no algorithm fixes, and re-recording is the better answer.
- The session, if you have it — MIDI parts beat audio every time
- A DAW with a modern pitch-shift algorithm, or a dedicated tool such as Melodyne
- The current key and the target key
- Somewhere to A/B against the original
Write down the current key and the target, and count the semitones by the shortest route. Moving from C to A is three semitones down, not nine up. Same destination, very different amount of processing.
If the reason for the change is a vocal range problem, try both directions. Down three and up nine put the singer in the same pitch class but completely different tessitura.
Select the MIDI clips and transpose. Then check two things that transposition breaks:
- Range. A bassline transposed down five semitones may drop below the instrument's usable range and turn to mud. Move it up an octave.
- Sampled instruments. Multisampled libraries change character across their range. A piano transposed a fifth may cross a sample zone boundary and sound noticeably different.
Turn formant preservation on before you do anything else. Almost every modern DAW has it, sometimes labelled formant lock or gender.
Then choose the algorithm. Most DAWs offer several:
| Material | Algorithm to choose |
|---|
| Solo vocal | Monophonic or vocal-specific mode, formants preserved |
| Bass or lead line | Monophonic mode |
| Full mix or chords | Polyphonic or complex mode |
| Drums and percussion | Do not pitch shift — replace or resample instead |
Listen for these, in this order:
- Vowel character. If the singer suddenly sounds younger or older, the formants moved. Turn preservation on or raise its amount.
- Hollowness. A thin, phasey quality on sustained notes means the algorithm is struggling. Try a different mode.
- Transient smearing. Consonants and picked attacks losing their edge.
- Acoustic body. A shifted acoustic guitar can still be perfectly clean and simply sound like a smaller or larger guitar. That is not an artifact, it is physics — but it may not be what you want.
Re-record when:
- The shift is more than about four semitones on a lead vocal
- The part is the focal point of the song and the artifacts are audible in context
- The instrument is acoustic and the change in body is obvious
- You have the performer available and the part is not difficult
Pitch shifting a whole mixed master is the hardest case of all, because you are shifting drums along with everything else. If you must, keep it to a semitone or two and accept some smearing on the cymbals.
A key change is not only a pitch change. It moves every part relative to the frequency space of the mix and to the instruments' sweet spots.
After transposing, check the low end — a bassline that sat perfectly at 55Hz now sits at 65Hz and may need the EQ moved. Check that the vocal chorus still lands in the singer's power range rather than just below it. And check that any resonance you tuned to the old key is not now fighting the new one.
If your source is a mixed audio file and you need a big key change, separating it into stems first and shifting each stem independently produces a better result than shifting the mix as one. Drums can be excluded from the shift entirely, and each remaining stem can get the algorithm that suits it. Tools that combine separation with a timeline — Veena, RipX — make that a single workflow rather than three.
Related reading: how to change tempo without artifacts, how to split stems from a song, and music theory basics with no theory.
Start making music in Veena
Free, browser-based, no downloads required.
Try Veena Free