vibevoice-tts

/**
 * @file preparer.ts
 * @description Podcast script preparation pipeline for VibeVoice TTS.
 *
 * Handles the full preparation workflow:
 *   - Abbreviation expansion (first-use or all)
 *   - Direction tag extraction ([emphasis], [slower], etc.)
 *   - Pause marker conversion ([pause 1s] → empty lines)
 *   - Sentence splitting & paragraph breaking
 *   - Script chunking (~500 words per chunk)
 *   - TTS parameter recommendations
 *
 * VibeVoice specifics:
 *   - Empty lines = natural pauses
 *   - No SSML support — plain text only
 *   - Max ~500 words per chunk for best quality
 *   - 150-170 WPM is standard podcast speaking rate
 */
import type { ScriptChunk, TTSRecommendations, PauseMarker, PrepareResult, DirectionTag } from "./types";
/**
 * Expand common abbreviations for TTS clarity.
 * `firstUseOnly`: expand only the first occurrence of each abbreviation.
 * `all`: expand every occurrence.
 */
export declare function expandAbbreviations(text: string, options?: {
    firstUseOnly?: boolean;
    all?: boolean;
}): string;
/**
 * Extract direction tags from text and return them as user-facing hints.
 * Unlike emotion annotations, direction tags are preserved as metadata
 * to guide the human operator on TTS parameter adjustments.
 */
export declare function extractDirectionTags(text: string): {
    cleanedText: string;
    directions: Array<{
        tag: DirectionTag;
        position: number;
        context: string;
    }>;
};
/**
 * Convert explicit [pause Xs] markers to empty lines.
 * VibeVoice interprets empty lines as natural pauses.
 */
export declare function convertPauseMarkers(text: string): {
    convertedText: string;
    pauses: PauseMarker[];
};
/**
 * Convert bracket-based expression markers to VibeVoice expression tags.
 * E.g., `[laughs]` → `<laughs>`, `[sighs]` → `<sighs>`.
 *
 * Supports ANY expression — not just known tags. The model can invent
 * new ones freely: `[roar]` → `<roar>`, `[sing softly]` → `<sing softly>`.
 */
export declare function convertExpressionBrackets(text: string): {
    convertedText: string;
    expressions: Array<{
        tag: string;
        position: number;
        original: string;
    }>;
};
/**
 * Split a paragraph into sentences.
 * Handles common edge cases: abbreviations, decimal numbers, ellipsis.
 */
export declare function splitIntoSentences(paragraph: string): string[];
/**
 * Split long paragraphs into smaller chunks of max N sentences.
 * Returns an array of shorter paragraphs.
 */
export declare function splitParagraphs(text: string, maxSentences?: number): string[];
/**
 * Count words in text (handles multi-language).
 */
export declare function countWords(text: string): number;
/**
 * Split a prepared script into chunks of approximately `targetWords` words.
 * Chunks are aligned with paragraph boundaries when possible.
 */
export declare function chunkScript(text: string, targetWords?: number): ScriptChunk[];
/**
 * Generate TTS parameter recommendations based on script analysis.
 */
export declare function generateTTSRecommendations(text: string, chunkSize?: number): TTSRecommendations;
/**
 * Run the full podcast script preparation pipeline.
 *
 * Steps:
 *   1. Expand abbreviations
 *   2. Extract direction tags (preserve as metadata)
 *   3. Convert pause markers to empty lines
 *   4. Split long paragraphs
 *   5. Chunk the script
 *   6. Generate TTS recommendations
 */
export declare function prepareScript(text: string, params?: {
    chunkSize?: number;
    paragraphPauses?: boolean;
    sentencePauses?: boolean;
    maxSentencesPerParagraph?: number;
    expandAbbreviations?: boolean;
    preserveDirections?: boolean;
    output?: "plain" | "script" | "json";
}): PrepareResult;
//# sourceMappingURL=preparer.d.ts.map