Vocals, delivery & harmony

Lineup + delivery + harmony

Three levers you can combine

Combine when useful:

• Lineup: solo, male-female duet, group unison
• Delivery: breathy falsetto, belted chorus, close-mic whisper, spoken-word verses
• Harmony: intertwining harmonies, call-and-response, octave doubles

Vague vs specific

Praise vs performance recipe

Copy-ready snippets

Vocal stack lines

Pillar 4 · Vocal in Prompt Builder.

Try it

Align lyrics and takes

Prompt Builder · If lyrics fight the vocal recipe, see Lyrics vs Style and use lyrics diff. For ready-made vocal-flavored lines, browse the Style library.

Going deeper

Performance language and retries

Gender or timbre hints are more useful when paired with range, delivery, and role rather than stereotypes. If vocals feel wrong, check Troubleshooting and compare equal baseline and variant samples; no fixed number of takes proves the cause.

How to test vocal wording

Observable cues, not undocumented internals

Suno does not document an internal voice-selection pipeline that lets us promise how a phrase is implemented. Treat words such as "breathy falsetto" and "belted chorus" as audible directions to test: keep the rest of the prompt stable, compare several variants, and score registration, intensity, distance, and layering.

Describe the performance you want instead of using praise such as "good singer" or "professional vocals." For example, compare "male-female duet, conversational intimacy" with "male-female duet, belted harmonies" and record which differences are consistently audible; do not assume that every word controls an independent internal setting.

The three levers of vocal control

Lineup, delivery, and harmony

Use three editorial categories to organize a comparison: lineup (solo, duet, group, choir), delivery (whispered, belted, spoken, melismatic), and harmony (unison, harmonized, call-and-response, layered). These are a writing framework, not a claim that Suno exposes three independent controls.

You do not need to fill every category. Start with the smallest description that captures the target, then add one missing dimension. A longer phrase such as "male-female duet, conversational verses, belted chorus, tight chorus harmonies" is useful only if you can evaluate each requested trait.

Vocal combinations to test

Starting points, not benchmark results

• Intimate solo: "close-mic solo vocal, breathy, whispered verses, gentle chorus": listen for vocal distance and verse/chorus contrast.
• Power duet: "male-female duet, belted chorus, intertwining harmonies, call-and-response verses": listen for two distinguishable roles and call-and-response phrasing.
• Choral: "mixed choir, four-part harmony, reverent, sustained pads": listen for ensemble density and sustained harmony.
• Rap delivery: "rhythmic vocal delivery, fast flow, double-time sections, ad-libs": listen for cadence and background responses.
• Spoken word: "spoken-word verses, sung chorus, narrative delivery": listen for delivery contrast between sections.
• Folk harmonies: "two-part harmony, stripped-back, acoustic, storytelling delivery": listen for a second vocal line without assuming exact voicing.

When vocals go wrong

Diagnosing and fixing vocal issues

Begin diagnosis with three prompt-side checks: replace vague praise with an audible delivery word, separate competing directions such as "whispered" and "belted," and check whether the lyric tone supports the requested performance. These checks narrow the experiment but do not prove that every vocal problem originates in the prompt.

If vocals stay off-target, simplify to one lineup word and one delivery word, generate equal samples for the baseline and variant, and add complexity one cue at a time. Persistent differences are evidence for the next test, not proof of a single internal cause.

Advanced vocal techniques

Beyond basic delivery tags

• Vocal effects: Compare texture requests such as "telephone vocal," "distorted," "vocoded," or "robotic" with an untreated baseline. They describe a target character, not an exact signal chain.
• Dynamic vocal arcs: Test labels such as [Verse - whispered] and [Chorus - belted] alongside a consistent Style request. They can help state the intended contrast but do not provide deterministic per-section control.
• Ad-libs and layers: Test "(ad-libs)" or "[Ad-libs]" around a short passage, then check whether background responses appear without obscuring the lead. The label is a request, not a guaranteed layer.
• Voice description: If a gender label matters to the arrangement, combine it with range and delivery words, then verify the audible result. Avoid treating gender labels as exact or stereotype-free controls.