AI audiobooks stumble on polyphonic characters; LLM pre-annotation fixes TTS output
dotey · x · 2026-10-22
Replying to a parent asking about making audiobooks for their kid, dotey argues current AI narration still falls short of professional human readers, and polyphonic pronunciation errors can actively mislead children—so pick human narrated versions when available.
For DIY AI audiobooks he shares a practical workflow: run the text through an LLM first, have an agent drive the TTS with pinyin annotations for polyphonic characters, pause positions, and emotion marks, then generate—the result improves noticeably.
He adds that paid TTS generally beats open source, though he uses Gemini TTS out of habit rather than price-performance claims.
More from coding & agent
- Nous Research ships official Hermes Agent plugin for Meshy AI 3D generation via MCP — NousResearch · 2026-10-22
- Support AI Regurgitating Stale Docs Is a Growing Failure Mode — Give It Real Platform View — joshwhiton · 2026-10-22
- CallBaba launches BabaBench, a benchmark testing AI agents on real patient advocacy tasks — mattturck · 2026-10-22
- 70%-95% failure rates: devs say AI agents still too unreliable for real workflows — Luke17man · 2026-10-22
- How do you actually test agents before production? Devs tackle failure-path testing — Last-Bluejay-4443 · 2026-10-22
- A Reddit long-read proposes the Risk and Judgement pattern for bounding agent autonomy — mostly_deterministic · 2026-10-22