How Much of Vocal Editing Preparation Can Be Automated?
Structural processing across regions, timing, and level
The cost of preparing a vocal recording does not come only from difficult edits. It comes from repeating many kinds of decisions across many regions: finding silent passages, distinguishing breaths from singing, fixing boundaries, reviewing timing, and examining differences in input level between phrases.
The central design question is not simply what can be automated. It is also what should not be automatically committed. Musical expression cannot be reduced to measurements of a waveform. Machines can handle observation and repeated mechanical work without taking authority over expressive choices.
ShushuLab therefore treats this stage as structural processing: preparation of the source material before tonal shaping or a finished mix.
1. Start with the structure of the recording
A vocal recording may contain distinct classes of regions:
- Silence: passages without vocalization
- Breath: breathing-related passages
- Singing: sung material
- Unknown: material whose classification should remain unresolved
Reducing these categories to a simple sound-versus-silence decision can remove an intended breath or disrupt the relationship between breathing and an attack. Classification is more useful as a reviewable editing proposal than as an unquestionable answer. An Unknown class gives the engineer room to resolve uncertain cases.
Muting silence and shortening the timeline are also different operations. When a recording is aligned to an accompaniment, a silent passage may need to remain in its original temporal position. Preserving that structure is a prerequisite for dependable downstream editing.
2. Timing correction is not indiscriminate grid alignment
The attack of a sung phrase, the breath preceding it, and the connection to neighboring regions all matter.
Moving an attack raises further questions: should its breath follow, which regions may move, and where must propagation stop? A local correction that shifts every subsequent region can damage material outside the intended edit.
The structural objective is therefore to create timing proposals with bounded movement and clear scope, then let the engineer judge whether they remain musically natural.
Numerical alignment and a convincing performance are not identical.
3. Level preparation is not a substitute for compression
Large differences in source input level can change how later dynamics or saturation processors respond. Yet forcing every phrase to the same apparent loudness is not the answer. Intended differences between sections and expressive movement within phrases belong to the performance.
The relevant structural task is to identify and, where appropriate, address large differences in input conditions between material, phrases, or sections, without erasing intended contrast. This is separate from setting final loudness or building the finished vocal tone.
In the ShushuLab workflow, Vocal Prep is responsible for upstream structure, while Formant Preamp addresses downstream conditioning and standardization. The stages have distinct responsibilities rather than interchangeable feature lists.
4. Analysis proposes; the engineer confirms
A reviewable workflow separates observation from commitment:
- Analyze the state of regions, boundaries, timing, and input level.
- Proposal creates candidates without finalizing the result.
- Preview lets the engineer evaluate the candidates and find exceptions.
- Manual Correction adjusts only the classifications, boundaries, timing, or gain decisions that require intervention.
- Confirm / Bounce turns the reviewed work into material for the next stage.
This approach transfers repetitive work to the software while preserving decision authority. A high automation rate means little if an incorrect proposal cannot be inspected, reversed, or corrected.
5. The output of preparation is not a finished vocal sound
Structural preparation is not pitch correction, formant processing, tonal EQ, compression, saturation, or spatial mixing. Those tasks retain different goals and require different decisions.
The intended result is material whose region structure is easier to understand and whose repeated preparation burden can be reduced. The extent of any downstream time saving depends on the source, settings, and workflow and must be measured rather than assumed. A design objective is not the same as a verified performance claim.
Vocal Prep in practice
Vocal Prep is ShushuLab's implementation of this structural approach. Its Windows VST3 has been checked as an ARA Event FX in Studio One 7 Professional. It follows a workflow in which analysis produces reviewable proposals, exceptions can be corrected manually, and the engineer confirms the result before output.
This article explains the design principles and division of responsibilities. For the features, instructions, and limitations of the currently sold version, refer to the product's official guide.