Audiobook Tools
The hard part of an audiobook is not the voice but consistent pronunciation and clean chapter delivery.
Tool profile
- Kind
- service
- Vendor
- ElevenLabs、Microsoft Azure AI Speech
- Form
- 服务
- Platforms
- 浏览器 · macOS · Windows
- Price
- 按量 + 订阅
- Jobs
- 有声书 · 旁白 · 本地化
- Verified
- 2026-10-31
Related
Cut long text by chapter
Paste a hundred thousand words at once and the engine truncates, times out or starts over. Cut by chapter or scene into blocks of one to three thousand characters, each generated, auditioned and re-recorded on its own. Mark paragraph pauses instead of relying on blank lines, and spell out numbers, units and dates the way they should be read. Generate dialogue and narration as separate tracks so level and tone can be adjusted later.
The pronunciation lexicon decides the result
- List names, places and brands first, and add entries in order of appearance.
- Write heteronyms by context instead of trusting the default reading.
- Spell numbers and symbols as speech: years, decimals, percentages.
- Specify whether an abbreviation is spelled out or read as a word.
- Choose IPA or respelling, and keep one convention across the team.
- Respect case and inflections; add plurals and possessives separately.
- After each lexicon change, spot-check three places for collateral damage.
Chapter delivery checklist
- One file per chapter, named with book, chapter and version.
- Half a second of silence at each end; no clipped first or last word.
- Consistent loudness across the book, not rescued by compression later.
- One sample rate and bit depth, in the format the platform asks for.
- Title, author and narrator complete in the metadata.
- Audition the first and last sentences of every chapter; check proper nouns one by one.
- Keep an untreated dry master so a revision does not regenerate whole chapters.
What not to hand to a synthetic voice
Poetry, dialect dialogue, a sob in a memoir, legal disclaimers: synthetic speech struggles to carry them, and a children's book that needs character acting still needs a human to set the tone. Voice cloning requires written consent from the person, and public release must disclose synthetic speech where the platform requires it. The licence covering training and reference audio decides whether the result can be sold.
Start with one chapter
Make a fifteen-minute sample chapter, settle the lexicon, pauses and loudness, then generate in batches. After that, read loudness normalisation, synthetic-voice disclosure and copyright ownership.
Listen in the library
All tracks from midicn-lib — inline badge is the license tier (C1 commercial / C2 non-commercial / C3 study only).
Along the same line
Sources
- ElevenLabs、Microsoft Azure AI Speech 与 Amazon Polly 官方文档中的长文本、SSML 与发音词典说明。
- W3C Speech Synthesis Markup Language 规范中停顿、重音与发音标注。
- Audible 与 Spotify 有声书投稿规范中的章节、响度与元数据要求。
- 合成语音披露与声音克隆同意的平台政策。