Produce an Audiobook Episode
Chapter the text first; casting and breath decide whether a listener stays.
Tool profile
- Kind
- guide
- Vendor
- —
- Form
- 案例
- Platforms
- 通用
- Price
- 不适用
- Jobs
- 有声书 · 长文本朗读 · 章节交付
- Verified
- 2026-10-31
Related
What leaves the studio
A thirty-minute episode runs to eight or ten thousand Chinese characters. The platform receives one file per chapter plus a full-episode master, a chapter list, a pronunciation sheet and a QC log. The technical limits are fixed: noise floor below -60 dB, RMS between -23 and -18 dB, peaks no higher than -3 dB. Leave half a second to a second of silence at the head and tail, and keep that gap consistent between chapters. Listeners hear half an hour a day, so one sibilant or plosive gets magnified on a commute.
Read, record, chapter
Read the whole text first and cut chapters at scene changes. Build a pronunciation sheet for names, dialect words and foreign terms, and note the register and pace of each character. Record the narration straight; characters get a light change of timbre, not three accents from one throat. Export chapter by chapter, set loudness once, and audition each file for sibilance, mouth noise and sagging phrase endings. Generation is only for five-to-eight-second transitions between chapters, never for the body of the read.
Where the tools sit
Record and edit in Audacity, Adobe Audition or Descript; Descript's text editing is good for fixing fluffs, and its pronunciation lexicon locks names. iZotope RX handles sibilance, room reflection and noise floor, while Auphonic or Ozone matches chapter loudness. Make transition beds with Suno or Udio, instrumental and short. Keep the chapter list and pronunciation sheet in the delivery folder, and put the chapter number in every file name.
Three listening passes
| Check | Pass line | What the ear hears |
|---|---|---|
| Noise floor | Below -60 dB | No hum or room tone in silence |
| Loudness | -23 to -18 dB RMS | No jump between consecutive chapters |
| Peaks | No higher than -3 dB | Emphatic lines do not clip |
| Chapters | Even head and tail silence | Playback does not lurch |
| Characters | Audibly different timbre | You can tell who speaks with eyes closed |
| Pronunciation | Matches the sheet | Names need no re-read |
Where it breaks
A character is overplayed and performance outranks content. Sibilance is over-repaired and the voice turns dull. Each chapter is limited on its own, so the book gets louder chapter by chapter. A transition bed carries a hummed vocal that fights the narrator. The pronunciation sheet is not enforced and one name gets three readings. The licence covers background music only, not commercial audiobook release and sample clips.
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
- ACX 公开的音频提交要求，涉及底噪、RMS 与峰值。
- 有声书朗读与章节交付的制作惯例。