AdherenceHow strictly the model follows your style instruction. • Lower (1.5–1.8): more variety and emotional spark, occasionally drifts. Best for dramatic, expressive reads. • Default (2.0): balanced. • Higher (2.2–3.0): rigidly disciplined, very stable, less surprise. Best for repeatable brand voice. Tip: at 1.6 the "Angry" style might genuinely shout; at 2.5 it stays controlled.

Quality vs SpeedHow many refinement passes the model makes per audio clip. • Lower (4–8): faster generation, slightly rougher prosody. • Default (10): balanced. • Higher (15–25): smoother, more natural pacing. Best for production clips. Think of it like draft vs final render in video software. For Hindi or emotional reads, 15+ helps noticeably.

Ambient BackgroundLayer a low-volume ambient track under the speech to make the recording feel like it's happening in a real environment. • None — clean studio recording (default for ads, audiobooks). • Call Center — soft typing, distant chatter; for support flows. • Office — gentle pink-noise hum; corporate calls. • Street — outdoor brown-noise rumble; door-to-door, field reporting. • Crowd — distant murmur; small gatherings. • Rally — louder crowd presence; political stadium speeches. Mixed server-side at -14 to -28 dB, looped to match the speech length.

Text CleanupPre-processes your text before synthesis: numbers ("123" → "one hundred twenty three"), expanded abbreviations, normalized punctuation. Leave on for almost everything. Turn off only if you need exact verbatim pronunciation of every character (e.g., reading out a code or model number literally).