import type ASRGgml from '@qvac/asr-ggml';
import type TranslationNmtcpp from '@qvac/translation-nmtcpp';
export interface LlmStats {
    TTFT?: number;
    TPS?: number;
    ppTPS?: number;
    CacheTokens?: number;
    promptTokens?: number;
    generatedTokens?: number;
    avgConcurrentSeq?: number;
    toolDefinitionsDropped?: number;
    backendDevice?: 'cpu' | 'gpu';
    stopReason?: 'none' | 'eos' | 'antiprompt' | 'predictionLimit' | 'sequenceLimit' | 'contextOverflow';
}
export interface LlmResponse {
    stats?: LlmStats;
    iterate(): AsyncIterable<string>;
}
export type NmtStats = TranslationNmtcpp.RuntimeStats;
export type NmtResponse = TranslationNmtcpp.TranslationResponse;
/**
 * @qvac/tts-ggml `RuntimeStats`. Restated here rather than imported from the
 * addon so a typecheck does not depend on which addon version happens to be
 * installed; keep it in step with the addon's `RuntimeStats`.
 */
export interface TtsStats {
    audioDurationMs?: number;
    totalTime?: number;
    realTimeFactor?: number;
    tokensPerSecond?: number;
    totalSamples?: number;
    /** Audio8 and MOSS only: codec frames generated (Audio8 on a 46 ms grid, MOSS at 12.5/s). */
    generatedFrames?: number;
    backendDevice?: number;
    backendId?: number;
    gpuUnsupported?: number;
    enhancerBackendDevice?: number;
    enhancerBackendId?: number;
    /** Audio8 only, macOS / iOS: 1 while the Core ML codec sidecar is attached. */
    codecSidecarLoaded?: number;
    /** Audio8 only: 1 when the synthesis ran its codec on that sidecar, 0 on ggml. */
    codecOnCoreml?: number;
}
/**
 * @qvac/tts-ggml `TTSOutputChunk & SentenceStreamChunkMeta`. The chunker runs
 * for `runStream()` and for `run({ streamOutput: true })` alike, so the meta
 * fields are populated on both paths.
 */
export interface TtsOutputChunk {
    text?: string;
    outputArray: ArrayLike<number>;
    /** Rate of `outputArray`; moves with `outputSampleRate` and the enhancer. */
    sampleRate?: number;
    chunkIndex?: number;
    sentenceChunk?: string;
    /** True on the final pre-chunked output; undefined when it is not known up front. */
    isLast?: boolean;
}
export interface TtsResponse {
    stats?: TtsStats;
    iterate(): AsyncIterable<TtsOutputChunk>;
}
export interface EmbedStats {
    total_time_ms?: number;
    tokens_per_second?: number;
    total_tokens?: number;
    backendDevice?: 'cpu' | 'gpu';
    context_size?: number;
}
export interface EmbedResponse {
    stats?: EmbedStats;
    await(): Promise<Float32Array[][]>;
}
export type TranscribeStats = Partial<ASRGgml.WhisperRuntimeStats & ASRGgml.ParakeetRuntimeStats>;
export type TranscribeAddonSegment = ASRGgml.TranscriptionSegment;
export type TranscribeAddonVadEvent = ASRGgml.VadEvent;
export type TranscribeAddonEndOfTurnEvent = ASRGgml.EndOfTurnEvent;
export type TranscribeAddonOutput = ASRGgml.ASRStreamOutput;
export interface TranscribeResponse {
    stats?: TranscribeStats;
    iterate(): AsyncIterable<TranscribeAddonOutput>;
}