Practical fixes: split narration into short chunks (1–3 sentences) and synthesize per chunk; increase ElevenLabs stability and lower similarity_boost to reduce erratic sibilance; use SSML pauses and tighten punctuation to control breaths; post-process with a de-esser and automatic breath remover (Audacity or iZotope RX), apply gentle EQ/high-pass and light compression, then normalize and stitch segments for final mastering. See ElevenLabs docs for exact API parameter names.