@voxdown/produce
v0.1.0
Published
Turn a VoxDown story into an audiobook with the voice service you choose: a production script any tool can read, adapters (ElevenLabs, OpenAI-compatible speech such as Kokoro, Azure), and chapters assembled as WAV, MP3 and M4B.
Maintainers
Readme
@voxdown/produce
Make a VoxDown story into an audiobook with the voice service you choose: each character in a voice of their own, directions followed where the service can, the same pacing whatever the service, and chapters as WAV, MP3 and M4B.
npm install @voxdown/produce voxdownMost people use it through the command, npx -y @voxdown/cli produce story.vxd; docs/produce.md explains it.
import { parse } from "voxdown";
import { elevenlabs, productionScript, produce } from "@voxdown/produce";
import { folderStore, wavFile } from "@voxdown/produce/node";
const script = productionScript(parse(text));
const made = await produce(script, {
adapter: elevenlabs({ key: process.env.ELEVENLABS_API_KEY! }),
voices: { Narrator: "JBFqnCBsd6RMkjVDRZzb", Mother: "XrExE9yKIg1WjnnlVkGX" },
store: folderStore(".clips"),
output: (chapter) => wavFile(`chapter-${chapter.number}.wav`, 24000),
});
// made.chapters[i].cues: when each line, and each word, is heard.productionScript(story): everything heard, as plain data (ids, chapters, scenes, pauses, what follows each line).- Adapters:
elevenlabs,openaiSpeech(OpenAI-compatible servers: Kokoro, LocalAI, vLLM-Omni, OpenAI; seePRESETS) andazure. Writing your own takes one function: see Writing an adapter. estimate(script, …): characters, requests and cost, before anything is made.castVoices(people, voices): a first casting, a voice for each character.chapterSsml(script, chapter, voices): a chapter as SSML, for tools that take it.@voxdown/produce/node: clips kept on disk (folderStore), chapters as WAV (wavFile), and with ffmpeg, MP3 chapters (chapterMp3) and an M4B with its chapters marked (bookM4b).
The core (productionScript, the adapters, produce) only uses fetch, so it runs in browsers too.
License
MIT
