This is the holy grail question — and there's finally a real answer plus the classic workaround.
1. Use Personas (the real fix). Suno's persona feature lets you save a voice from a generation and apply it to new songs. When you get magic, save it IMMEDIATELY — don't assume you'll recreate it.
2. The classic workaround: the exact prompt recipe. Copy the FULL style box verbatim, note the model version, and reuse both. Voice = style wording + model version. Change one comma and the voice drifts.
3. Describe the voice like a casting director: warm raspy female alto, slight country lilt, conversational — the more specific the vocal description, the more reproducible the voice.
4. Keep a voice journal. For each keeper voice: the song name, the exact style text, the model version, and what makes it special. This is your casting book.
5. Accept 90%. Even with personas and recipes, the 'same' voice varies slightly between songs — like a singer on different days. 90% consistent is a win; chase 100% and you'll burn credits forever.
Good luck! Hank founder of SonorixPG