有聲書自己做 · DIY Audiobook — 說一句話，做出一本有聲書

示範Demos

三種故事，同一句話開始Three kinds of story, one sentence each

全部用內建的免費語音做的，沒有用任何付費服務。All made with the free built-in voices — no paid service involved.

兒童故事Children's story

〈獅子與老鼠〉The Lion and the Mouse

溫暖的說書人聲音、5張本機生成的插圖、自動生成的音樂盒搖籃曲。3分鐘。A warm storyteller, five illustrations drawn locally, a generated music-box lullaby. 3 minutes.

播放影片 ↑Play the video ↑

紅髮會bedtime radio drama

睡前廣播劇Bedtime radio drama

福爾摩斯〈紅髮會〉“The Red-Headed League” (Mandarin)

從英文原著翻成台灣中文，旁白、福爾摩斯、委託人各一個聲音。前90秒。Translated from the original into Taiwanese Mandarin; narrator, Holmes and the client each with their own voice. First 90 s.

The Speckled Bandaudiobook · English

English

“The Speckled Band”

英文有聲書，英國腔旁白與福爾摩斯。前90秒。English audiobook with British narrator and Holmes. First 90 s.

怎麼做How it works

你說一句話之後，發生了什麼What happens after your one sentence

Claude負責語言模型擅長的事；程式負責語言模型不擅長的事。任何要花錢的服務，都要先通過檢查才會被呼叫。Claude does what language models are good at; scripts do what they are bad at. Nothing that costs money runs until the checks pass.

同一隻獅子、同一隻老鼠：每張圖都用一模一樣的文字描述角色，畫圖模型才不會每張畫得不一樣。The same lion, the same mouse: every prompt describes each character in identical words, so the pictures stay consistent.

抓原文Fetch the textscript

Project Gutenberg、網址或你自己的檔案，自動切成段落。Project Gutenberg, a URL or your own file, split into paragraphs.

改寫成用聽的Adapt for the earClaude

逐段翻譯或改寫，一段對一段，之後再對照原文審一次。Translate or adapt paragraph by paragraph, then review against the original.

選角Cast the voicesClaude

判斷每一句是誰說的，替每個角色挑聲音；也寫好插圖和配樂的提示詞。Decide who says every line and pick a voice for each character; write picture and music prompts.

花錢前先檢查Check before spendingscript

漏翻的段落、說話者數不對、兩個角色撞聲……都在這裡被擋下來。Skipped paragraphs, miscounted speakers, two roles on one voice — blocked here.

配音、配樂、插圖Voices, music, picturesscript

每個聲音自動對齊旁白音量；配樂先檢查有沒有卡在同一個和弦。Every voice levelled to the narrator; music checked for frozen chords first.

混音與成品Mix and finishscript

−14 LUFS母帶、影片、封面、逐句字幕、章節和上架文案。−14 LUFS master, video, cover, sentence-level subtitles, chapters and upload text.

可以選Options

從免費開始，想要更好再升級Start free, upgrade when you want

全部有一步一步的教學：怎麼申請key、怎麼用Suno做配樂、怎麼在自己電腦上畫圖。Every option has a step-by-step guide: getting a key, making music with Suno, drawing pictures on your own computer.

	免費、不用設定Free, zero setup	要帳號或keyWith an account or key	自己的Your own
語音Voices	微軟Edge語音Microsoft Edge voices	MiniMax（中文最自然）、OpenAIMiniMax (best Mandarin), OpenAI	—
配樂Music	自動生成的音樂盒搖籃曲Generated music-box lullaby	Suno（附教學與提示詞範本）Suno (guide + prompt templates)	任何音檔Any audio file
插圖Pictures	本機ComfyUI（免費，要裝一次）Local ComfyUI (free, install once)	OpenAI、MiniMax	任何圖、任何app畫的Any picture, from any app
風格Style	兒童故事 · 睡前廣播劇 · 有聲書Children's story · bedtime radio drama · audiobook

用耳朵調出來的預設值Defaults tuned by ear

將近兩百集踩過的坑，直接內建Nearly two hundred episodes of lessons, built in

不用懂混音。這些數字都是在真正上架的節目裡，一集一集聽出來的。No mixing knowledge needed. Every number here was tuned on episodes that actually went out.

0.35s

對白和「他說」之間的停頓。語音引擎每段前後都會墊靜音，不修掉會隔整整一秒。Between a line and its “he said”. TTS pads every clip with silence; untrimmed, they sit a full second apart.

9dB

不同語音之間最多差這麼多。每個聲音都自動對齊旁白，半夜不用調音量。How far apart TTS voices can be. Every voice is matched to the narrator — no reaching for the volume at 1 a.m.

16.5LU

配樂比人聲低這麼多，有人說話時再自動壓下去。太大聲會吵，太小聲等於沒有。Music sits this far under the voice, and dips further while someone speaks.

−14LUFS

成品響度，YouTube、Spotify、Apple都能直接上。峰值量的是編碼後的檔案。Finished loudness, ready for YouTube, Spotify and Apple. Peaks measured on the encoded file.

開始Get started

五分鐘裝好，然後說一句話Five minutes to install, then one sentence

需要Claude Code、Python 3.9以上和ffmpeg。最後一行會告訴你的電腦現在能做什麼。You need Claude Code, Python 3.9+ and ffmpeg. The last line tells you what your computer can do right now.

git clone https://github.com/mickyyang-1407/diy-audiobook ~/.claude/skills/diy-audiobook
pip install -r ~/.claude/skills/diy-audiobook/requirements.txt
brew install ffmpeg # or: apt install ffmpeg / winget install ffmpeg
python3 ~/.claude/skills/diy-audiobook/scripts/doctor.py

把〈小紅帽〉做成給孩子聽的有聲書，要有插圖和輕音樂。Make “Little Red Riding Hood” a children's audiobook, with pictures and gentle music.

把〈紅髮會〉做成睡前廣播劇，慢一點、輕一點。Make “The Red-Headed League” a bedtime radio drama — slow and soft.

把〈龜兔賽跑〉做成十分鐘的日文兒童故事。Make “The Tortoise and the Hare” a ten-minute children's story in Japanese.

把〈花斑帶〉做成有聲書，每個角色不同聲音，不要配樂。Make “The Speckled Band” an audiobook, one voice per character, no music.

教學：Guides: 付費語音與API keypaid voices & API keys · 配樂與Sunomusic & Suno · 插圖：本機或雲端pictures: local or cloud · 為什麼預設值是這樣why the defaults