vibevoice-tts

{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "$id": "https://primary-data.dev/schemas/vibevoice-tts/formatting.json",
  "title": "VibeVoice TTS Format Options",
  "description": "Describes the formatting options accepted by the vibevoice-tts format tool. These options control text cleaning, multi-speaker script formatting, and voice/emotion annotations for VibeVoice TTS. VibeVoice does NOT support SSML — output is plain text in script format.",
  "type": "object",
  "properties": {
    "text": {
      "type": "string",
      "description": "The input text to format for VibeVoice TTS. Supports plain text, 'Speaker N:' prefix format for multi-speaker, or a JSON array of {speaker, text} objects.",
      "minLength": 1
    },
    "mode": {
      "type": "string",
      "description": "Speaker mode: 'single' (one speaker), 'multi' (parse Speaker N: prefixes), or 'auto' (detect automatically, fallback to single).",
      "enum": ["single", "multi", "auto"],
      "default": "auto"
    },
    "speakers": {
      "type": "integer",
      "description": "Number of speakers (1-4). Used for voice/emotion array indexing.",
      "minimum": 1,
      "maximum": 4,
      "default": 1
    },
    "voices": {
      "type": "array",
      "description": "Voice names per speaker, e.g. ['Alice', 'Frank']. Used as documentation for voice reference selection — the user picks matching audio samples for each VibeVoice speaker slot.",
      "items": {
        "type": "string"
      }
    },
    "emotions": {
      "type": "array",
      "description": "Emotion labels per speaker for voice reference selection. VibeVoice does NOT support inline emotion tags — emotion comes from the reference audio sample selected for each speaker slot.",
      "items": {
        "type": "string",
        "enum": [
          "neutral",
          "happy",
          "sad",
          "angry",
          "excited",
          "calm",
          "thoughtful",
          "whisper",
          "serious",
          "energetic",
          "warm",
          "cold",
          "mysterious",
          "dramatic"
        ]
      }
    },
    "clean": {
      "type": "boolean",
      "description": "Apply VibeVoice text-cleaning pipeline. Strips smart quotes, Chinese quotation marks, emoji, HTML/XML tags, URLs, non-speech annotations, tabs, and excess whitespace.",
      "default": true
    },
    "output": {
      "type": "string",
      "description": "Output format: 'plain' (just cleaned text), 'script' (Speaker N: prefix format suitable for VibeVoice), or 'json' (structured array of {speaker, text} objects).",
      "enum": ["plain", "script", "json"],
      "default": "script"
    }
  },
  "required": ["text"],
  "additionalProperties": false
}