@kithjs/voice-router

The voice quality layer. Sentence-aware chunking, slang expansion, pronunciation overrides, emoji-to-emotion translation, VoiceCharacter profiles, and a text transform pipeline.

Install

bun add @kithjs/core @kithjs/voice-router

Core Class: VoiceRouter

import { VoiceRouter, DEFAULT_ENGLISH_SLANG, DEFAULT_LAUGH_TAGS } from "@kithjs/voice-router";

const voice = new VoiceRouter({
  runtime,          // any RuntimeAdapter
  slang: { ...DEFAULT_ENGLISH_SLANG, ...DEFAULT_LAUGH_TAGS },
  pronunciation: { "Kaori": "kah-oh-ree" },
  transforms: [myCustomTransform],
  personaMode: "hype",
});

await voice.speak("Hey! That was fire haha omg");

What VoiceRouter Does

1. Strips emojis and emits emotion_state events based on emoji sentiment
2. Chunks text at sentence boundaries (no mid-thought prosody resets)
3. Expands slang ("fr" -> "for real", "lol" -> "[laughs]")
4. Applies pronunciation overrides ("Kaori" -> "kah-oh-ree")
5. Runs custom text transforms (your cleanup/formatting logic)
6. Deduplicates identical sentences within a configurable window

Default Slang Dicts

Dict	Contents	Example
DEFAULT_ENGLISH_SLANG	Universal text-speak	btw -> by the way, omg -> oh my god
DEFAULT_GENZ_SLANG	Gen Z shorthand	fr -> for real, ngl -> not gonna lie
DEFAULT_BOARD_SPORTS_SLANG	Skate/snow/surf terms	fs -> frontside, kf -> kickflip
DEFAULT_LAUGH_TAGS	Laugh text to v3 audio tags	haha -> [laughs], hehe -> [giggles]

VoiceCharacter Profiles

Bundle voice settings + slang + pronunciation into a JSON file. See Voice Characters.

Text Transforms

Add custom preprocessing before TTS:

const cleanMarkdown = (text: string) => {
  return text.replace(/\*{1,3}([^*]+)\*{1,3}/g, '$1');
};

const voice = new VoiceRouter({
  runtime,
  transforms: [cleanMarkdown],
});