Build a Singer in Five Questions: The Stack Method That Actually Works

Most disappointing AI vocals aren't the engine's fault. They're five decent settings arguing with each other.

On Toxic Sound, a voice is built as a stack — five slots, always the same ones: Range (where the voice sits), Tone (its colour), Grit (the texture on the surface), Delivery (how the lines are performed), and Environment (the room the voice lives in). Change the stack, change the singer. And the finding that matters most: vocal identity follows the cues, not the genre. The same five choices give you recognisably the same singer over a doom folk track or a punk track. The cues are the person. The genre is just the band behind them.

Build in this order — not the order on the screen

1. Delivery first. What is this singer doing? Chanting, drawling, shouting with a gang, threatening down a phone line. Behaviour is identity, so it goes first — never range.

2. Environment second. Where is this person standing? Stone hall, club room, telephone line. Ask it early, because the room disciplines everything after it.

3. Range third. Now that you know the behaviour and the room, where does the voice need to sit to perform that behaviour in that room? A chant in a hall wants depth. A riot in a club wants a voice that cuts.

4. Tone fourth. What colour is the threat, the warmth, the grief? Tone is the emotion the listener names first.

5. Grit last, and with a light hand. How much damage is on the surface? Grit is seasoning. A pinch focuses a stack. A bucketful buries it.

The 4-second test

Before you spend credits, say the five labels out loud as one description of one human being:

"A power baritone with a grave-deep tone and gravel in his throat, drawling slowly in a stone hall."

If those words describe a person you can picture, the stack will sing. If they describe five settings, it won't. Every great stack passes this test. Every beige one fails it.

The two failure patterns

The beige stack. Every cue pleasant, every cue safe — smooth tone, clean grit, delicate delivery, clean room. Nothing is wrong, and that's exactly what's wrong: there's no friction anywhere. Rule: every stack gets one slot of friction — one grit with texture, one tone with an edge, one delivery with a decision in it. Swap just the grit to something lightly raspy and the same singer suddenly has a history.

The lying room. A huge choral voice built for a cathedral, placed in a lo-fi bedroom. The performance says cathedral; the room says closet — and the listener's brain believes the room. It reads as comedy when it was meant as ritual. The reverse lies too: an intimate whisper drowned in cavernous reverb announces what the voice is trying to hide. Rule: the room never lies. Environment isn't seasoning sprinkled at the end. It's a claim about where this person is, and the other four slots have to be able to be true in that place. When a stack sounds almost right and you can't say why, check the room first.

Keep a stack journal

Every stack you try: write the five labels, the genre frame, and one sentence on what the singer felt like. In a month you'll own a personal cast of voices nobody else has — built entirely from cues everybody else also has access to. The registry is shared. The combinations, and the taste to choose them, are yours.

That's the moat a solo maker can actually dig.