Voice Characters

A VoiceCharacter profile bundles all voice-layer settings for a companion into a single JSON file. It controls how the companion sounds — not what it says (that's your agent's job).

Character Profile Shape

{
  "voice": {
    "stability": 0.55,
    "similarityBoost": 0.82,
    "style": 0.45,
    "useSpeakerBoost": true,
    "speed": 1.05
  },
  "slang": {
    "fr": "for real",
    "ngl": "not gonna lie",
    "gg": "good game"
  },
  "pronunciation": {
    "Kaori": "kah-oh-ree",
    "sugoi": "soo-goy",
    "ganbare": "gahn-bah-ray"
  },
  "personaMode": "hype"
}

Voice Settings

These control the ElevenLabs TTS parameters:

Setting	Range	Effect
stability	0-1	Higher = more consistent, lower = more expressive
similarityBoost	0-1	How closely to match the original voice
style	0-1	Style exaggeration (0 = neutral, 1 = maximum)
useSpeakerBoost	bool	Enhances voice clarity
speed	0.5-2.0	Playback speed (1.0 = normal)

Tuning tips:

• For smooth, natural speech: stability: 0.5-0.6, style: 0.3-0.5
• For expressive, animated speech: stability: 0.25-0.4, style: 0.6-0.8
• For fast-paced hype character: add speed: 1.05-1.1

Slang Expansion

The slang dict maps abbreviations to their spoken form. TTS engines read "fr" as "eff are" — slang expansion fixes this.

Kith ships four default dicts you can compose:

import {
  DEFAULT_ENGLISH_SLANG,    // btw, fyi, omg, lol, etc.
  DEFAULT_GENZ_SLANG,       // fr, ngl, sus, bet, etc.
  DEFAULT_BOARD_SPORTS_SLANG, // fs, bs, kf, hf, etc.
  DEFAULT_LAUGH_TAGS,       // haha → [laughs], hehe → [giggles]
} from "@kithjs/voice-router";

const slang = {
  ...DEFAULT_ENGLISH_SLANG,
  ...DEFAULT_GENZ_SLANG,
  ...DEFAULT_LAUGH_TAGS,
  ...character.slang, // character's custom overrides win
};

:::tip Laugh Tags DEFAULT_LAUGH_TAGS converts "haha", "lol", "hehe" into ElevenLabs v3 audio tags ([laughs], [giggles]) that render as real laughter. This only works with eleven_v3 — on older models the tags are read aloud literally. :::

Pronunciation Overrides

For names, loanwords, or anything the TTS mispronounces:

{
  "pronunciation": {
    "Kaori": "kah-oh-ree",
    "Hokkaido": "hoh-kai-doh",
    "SSX": "S S X"
  }
}

Wiring It Up

import profile from "./my-character.json" with { type: "json" };
import { voiceCharacterToRuntimeConfig, type VoiceCharacter } from "@kithjs/voice-router";

const character = profile as VoiceCharacter;

// Voice settings go to the runtime
const runtime = new PipecatRuntime({
  config: {
    pipeline: "elevenlabs",
    apiKey: process.env.ELEVENLABS_API_KEY,
    voiceId: process.env.ELEVENLABS_VOICE_ID,
    ...voiceCharacterToRuntimeConfig(character),
  },
});

// Everything else goes to the router
const voice = new VoiceRouter({
  runtime,
  character,
  slang: { ...DEFAULT_ENGLISH_SLANG, ...character.slang },
});