/**
 * Demofly Whisper integration — extract word-level timestamps from TTS audio.
 *
 * Shells out to the `whisper` CLI (same subprocess pattern as ffmpeg).
 * Uses whisper-base model — TTS audio is clean, so a small model achieves
 * near-perfect accuracy while keeping runtime under 5s per scene.
 *
 * Fallback: If Whisper isn't installed, returns duration-only data and
 * prints a warning suggesting `pip install openai-whisper`.
 */
export interface WordTimestamp {
    word: string;
    startMs: number;
    endMs: number;
}
export interface PhraseTimestamp {
    text: string;
    startMs: number;
    endMs: number;
    beatId: string;
}
export interface SceneTimestamps {
    sceneId: string;
    audioFile: string;
    durationMs: number;
    words: WordTimestamp[];
    phrases: PhraseTimestamp[];
}
export interface TimestampsData {
    scenes: SceneTimestamps[];
}
/** Check if the whisper CLI is available. */
export declare function hasWhisper(): boolean;
/**
 * Extract timestamps from all audio files in a demo's audio/ directory.
 *
 * If Whisper is available, produces word-level and phrase-level timestamps.
 * If not, falls back to duration-only data (empty words/phrases arrays).
 *
 * @returns TimestampsData written to audio/timestamps.json
 */
export declare function extractTimestamps(projectDir: string): TimestampsData;
//# sourceMappingURL=timestamps.d.ts.map