Controlling the voice in Suno: gender, tone, and delivery that stick

To stop Suno swapping singers on you, pin the voice in two places at once: put a vocal anchor line at the very top of the Lyrics field, in brackets, naming gender, timbre, and delivery in English ([raspy male lead, close-mic, weary]), and echo the same descriptor inside the Style field. The Lyrics anchor tells the model who is standing at the mic before it sings a single word. The Style descriptor keeps that choice from sliding when the arrangement shifts underneath. Do only one of the two and the voice holds for a verse, then quietly becomes someone else in the chorus.

Here is why it drifts in the first place. Suno casts a fresh singer for every generation, and within one track it can recast at a section boundary, because nothing told it the person at the mic in the verse is the same person in the chorus. The anchor is that instruction, written down. Once you give it, the model stops rolling the dice on casting and starts performing the voice you described.

Where the anchor goes, and why the top of Lyrics

The first thing Suno reads in the Lyrics field sets the casting for the whole track. So the descriptor goes on line one, in square brackets, above [Intro] or [Verse 1]. Brackets matter: everything inside them is a direction and never gets sung, so the anchor shapes the voice without leaking a word into the song.

A bare [Verse 1] with no anchor hands Suno an open casting call. It will pick a competent voice, but you have no say in which one, and the next generation picks a different one. Put the descriptor first and the field reads like a session sheet handed to a singer before the take: this is who you are, now sing.

One descriptor in two fields is the whole trick. Style is where Suno composes the arrangement, and if the arrangement decides on a big anthemic chorus it will reach for a belting voice to match, overriding a lone anchor in Lyrics. Naming the voice in Style too keeps the arrangement and the casting agreeing with each other instead of fighting.

The three axes: gender, timbre, delivery

Describe a voice on three axes and it stops being a generic Suno singer and starts being a specific person. Pick one word from each axis and stack them.

Register and gender. female vocal, male lead, androgynous vocal, deep male bass, high female soprano, boy choir, child vocal, old man vocal. This is the axis Suno gets wrong most often when you leave it out, because it infers gender from the genre and the genre lies (more on that below).

Timbre, the grain of the voice. breathy, raspy, smoky, warm, nasal, smooth, gritty, airy, honeyed, gravelly, velvety, brittle. This is the axis that makes two generations sound like the same throat. breathy and raspy are opposites and the model hears the difference clearly.

Delivery, how the voice performs the line. conversational, belted, whispered, spoken word, crooning, staccato rap, melismatic runs, deadpan, intimate close-mic, preacher cadence. Delivery is the axis you will change per section, so anchor the resting delivery and escalate later.

Stack one from each and you get something like breathy female vocal, smoky, conversational. That is specific enough that the second take sounds like the same singer as the first. You can add a mic and room note when it matters: close-mic, distant, lo-fi telephone vocal, stadium reverb. Resist the urge to pile on ten adjectives. Three precise ones out-steer ten vague ones, and past four or five the model starts averaging them into mush.

A full worked prompt

A neo-soul ballad is a good test case, because soul lives and dies on the character of the lead vocal, so a vague anchor shows immediately. Here is the whole prompt across the three fields.

Style:

neo-soul ballad, 70 BPM, warm Rhodes, soft brushed drums,
upright bass, breathy female lead vocal, smoky and intimate,
analog warmth, vinyl crackle, close-mic vocal

Lyrics (anchor on line one, then section tags):

[breathy female vocal, smoky, close-mic, conversational delivery]
[Intro]
[Verse 1]
Coffee going cold on the windowsill again
You said you'd call by nine, it's almost ten
[Chorus]
And I keep the porch light on
Long after the reason's gone
[Verse 2]
Rain on the roof sounds a lot like your name
Every quiet night starts to feel the same
[Bridge: pulled back, near-whisper]
Maybe I was never waiting for you
[Chorus: same breathy female vocal, a touch fuller]
And I keep the porch light on
Long after the reason's gone
[Outro]

Exclude / Negative:

male vocal, choir, harsh belting, autotune, robotic vocal, spoken word

Read what each field is doing. Style and the anchor agree on breathy female so the arrangement does not reach for a belter. The bridge bracket drops the delivery to a near-whisper for one section without changing the singer. The second chorus repeats same breathy female vocal because a bare [Chorus] is exactly where casting tends to wander, and that one repeat pins it. The Exclude line shuts out the two most common intruders in soul: a phantom choir swelling behind the chorus, and a male vocal the model sometimes adds as a duet partner nobody asked for.

Whisper, falsetto, belt: controlling the energy of the voice

Whisper, falsetto, and belt are all delivery, and Suno responds to them cleanly when you put the word where it should act. A change meant for the whole song goes in the anchor and Style. A change meant for one section goes inside that section's bracket, and only there.

whispered or breathy whisper pulls the energy right down, intimate and close, good for a bridge or an opening line. falsetto or head voice lifts the voice above its chest range, thinner and more fragile, the classic lift into a hook. belted or full chest voice, powerful is the opposite pole, maximum energy, the voice pushed to the front of the chorus.

The practitioner mistake here is putting belt in the anchor when you only want the chorus belted. Do that and the whole song shouts, verses included, and you lose all the contrast that made the belt land. Anchor the resting voice at conversational, then escalate per section: [Chorus: belted, full chest, powerful]. The gap between a conversational verse and a belted chorus is where the drama lives. Flatten it and every line arrives at the same volume, which reads as tiring, not big.

Holding one voice across sections and extensions

The anchor holds the voice within a single generation. The moment you Extend a track, or roll a second version, Suno is free to recast, because a new generation is a new casting call unless you lock the voice down. That lock is a Persona.

A Persona captures the voice from a track you already like and reuses it, so every extension and every new song pulls from the same throat instead of a fresh one. Reach for it in three situations: a song longer than one generation can cover, a set of tracks meant to sound like one artist across a project, or any time an Extend came back with a stranger singing your second half. Setup details and limits shift as Suno updates, so check the current app for how personas are captured and applied rather than trusting a fixed click path.

A Persona is not a substitute for prompt craft, it works with it. The Persona carries the timbre and identity of the voice. Your Style and your anchor still steer the delivery and energy section by section, so you can keep the same singer and still move them from a whispered verse to a belted chorus. Think of the Persona as the singer and the prompt as the direction you give that singer on the day.

Duets and backing vocals

Two voices need two anchors and section tags that name who is singing where. Declare both voices up top, then hand each section to one of them:

[male lead, raspy and warm]
[female harmony, airy and high]
[Verse 1: male lead]
[Chorus: both, call and response]
[Bridge: female lead, close-mic]
[Final chorus: both in harmony]

Without the per-section tags, Suno blends the two into an averaged voice or lets one drown the other. Naming who owns each section keeps the trade clean and the duet readable.

Backing vocals are the other half of this. Left unspecified they either vanish or take over, so name their size and their place in the mix. subtle backing harmonies, gospel choir on the chorus only, stacked vocal harmonies low in the mix, single doubled lead, no choir. The phrase on the chorus only is doing real work there: it stops a choir you wanted for eight bars from padding the entire song.

Common mistakes

The voice changes between takes. No anchor, or an anchor only in Style and not in Lyrics. Fix: put the descriptor in both fields, and for anything spanning more than one generation, capture a Persona so the voice survives the Extend.

Wrong gender. You named the genre but not the singer, so Suno inferred gender from the genre and guessed. trap and drill skew male, pop ballad and bedroom pop often come back female. Fix: state male vocal or female vocal outright, near the front of Style and on the first line of Lyrics. Do not leave the singer's gender to the genre's stereotype.

The voice drifts right after a section tag. A bare [Chorus] can quietly reset the casting. Fix: repeat the key descriptor in the section bracket where it wanders, [Chorus: same raspy male lead]. You do not need it on every section, just the ones that slip.

Everything comes out over-belted. Energy words like powerful, anthemic, epic, soaring in the anchor push every single line to full voice. Fix: keep the verses conversational and reserve the big words for the chorus bracket alone.

Descriptors written in Russian or another language. Suno reads English vocal descriptors far better than translated ones. придыхательный женский gives you a worse and less predictable result than breathy female vocal. Keep every promptable word English; only the sung lyric lines follow the song's language.

FAQ

Why does the voice change halfway through my song? Because nothing told Suno the chorus singer is the verse singer, and a section boundary is where it feels free to recast. Put a vocal anchor on the first line of Lyrics and repeat the key descriptor in any section bracket where the voice wanders.

How do I force a specific gender? Write male vocal or female vocal explicitly, both near the front of the Style field and as the first bracketed line of Lyrics. Do not rely on the genre to imply it, since genre stereotypes guess wrong often enough to matter.

Can I keep the same singer across an Extend? Not from prompt words alone, because each generation recasts. Capture the voice as a Persona from the take you like, then apply that Persona to the extension and to any follow-up tracks that should sound like the same artist.

What is the difference between whisper and falsetto in tags? whispered lowers energy and breathiness while staying in the normal range, intimate and quiet. falsetto or head voice raises the pitch above the chest range, thin and fragile. They are different axes, so you can even combine them: [Bridge: falsetto, near-whisper].

How do I get two singers who trade lines? Declare two anchors at the top, one per voice, then tag each section with who sings it: [Verse 1: male lead], [Chorus: both, call and response]. Without the per-section ownership Suno averages the two into one voice.