Auto Shift: Choose Target Notes Before Correcting Pitch

This is a scale experiment, not a singer repair

Auto Shift estimates one fundamental pitch and shifts the audio toward an allowed note. The example is an original eight-second synthesized, vowel-like phrase with one note at a time. There is no human singer, lyric or voice model. Its repeating A, B, C and G fundamentals are already close to equal-tempered pitches; switching to Chromatic may therefore change very little.

Instead, the starting setting deliberately chooses C Penta Minor. That scale allows C, E-flat, F, G and B-flat. Some source notes lie outside that set, making the targeting decision more apparent than a tiny tuning adjustment. This is musical remapping, not evidence that an original performance was wrong. Bass is selected under Input because this phrase sits around 98–131 Hz; the choice sets the detector range, not a bass instrument sound.

Separate the destination from the journey

1. In Quantizer, keep Root at C and compare Penta Minor with Chromatic. Concentrate on the first two notes of the phrase; the C and G notes already belong to both sets.
2. Return to Penta Minor and reduce Correct from 100% to zero. You are reducing the pull toward the target, not choosing a different target. Intermediate amounts can leave a note between its original pitch and the allowed note.
3. Restore Correct, then raise Smooth from 15% toward 70%. Listen at note beginnings for a slower arrival rather than an immediate step. If the destination never feels settled during a short note, reduce Smooth.

The small keyboard edits allowed pitch classes; it does not play a MIDI instrument. Changing Root relocates the scale pattern. A named scale is a starting mask, not automatic key recognition or an understanding of the chord underneath. Choose the notes with the arrangement in mind and keep a meaningful set enabled. Do not remove a passing tone merely because it looks outside the main key.

Listen for the detector getting the wrong job

A chord, noisy attack or overlapping notes can make a single-pitch estimate ambiguous. High, Mid and Bass narrow the expected input range; none turns this into independent correction for every note of a polyphonic recording. If a local upload jumps unexpectedly, first try a clean, isolated monophonic passage and the appropriate Input range. More Correct is not a cure for an unstable estimate.

The processor uses spectral analysis, which adds latency and can soften transients. The clean synthetic example does not test breath, consonants, expressive scoops or performance bleed. Use it to learn the controls, then reassess a source you own at a comfortable level. A natural-sounding result on real singing is not guaranteed by this demonstration.

Add an effect only after the correction is understandable

On the Pitch tab, Pitch and Fine add manual transposition beyond the correction. Formant is measured in octaves and changes the spectral envelope, while F.Follow controls its relationship to pitch movement. Try small changes and listen for hollow or smeared tone. These are timbre controls, not a way to recover missing articulation or recreate a particular person.

Vib Rate and Vib Depth add periodic pitch movement. Leave depth at zero while diagnosing correction; otherwise deliberate modulation can sound like tracking trouble. The LFO tab adds separate pitch, formant, volume and pan modulation. Introduce one destination at a time. Finally compare Dry/Wet at 100% and a lower amount: mixing original and corrected notes creates a doubled texture, which may be attractive but is not simply gentler tuning.