import { Observable } from 'rxjs';
import type { CacheOptions } from './legacy-phrase';
import type { TtsHandle } from './tts-handle';
/**
 * How TTS audio is chunked and fed to the mixer.
 *
 * - **`sentence`**: split on sentence boundaries and start synthesis per sentence (default).
 * - **`streaming`**: incremental chunks as they arrive, only for streaming-capable vendors
 *   (for non-streaming vendors the runtime falls back to sentence handling).
 * - **`full`**: accumulate the whole input and synthesize it as one segment after the input completes.
 */
export type TtsStrategy = 'sentence' | 'streaming' | 'full';
/**
 * TTS vendor alias understood by ScriptEngine.
 *
 * Dedicated TTS connectors are selected by aliases such as **`"elevenlabs"`**,
 * **`"google"`**, and **`"voctiv"`**. The default TTS path can also accept
 * compatible aliases such as **`"azure"`** or **`"neuro_v3"`**, depending on
 * deployment configuration. The open string form keeps the SDK compatible with
 * deployment-specific aliases.
 */
export type TtsVendor = 'elevenlabs' | 'google' | 'azure' | 'yandex' | 'sber' | 'viettel' | 'voctiv' | 'neuro' | 'neuro_v2' | 'neuro_v3' | 'neuro_service' | 'async_azure' | 'stc' | (string & Record<never, never>);
/**
 * Options for {@link import('./media-channel').ChannelAudio.say},
 * {@link import('./media-channel').ChannelAudio.play}, and
 * {@link import('./media-channel').ChannelAudio.presay}.
 *
 * ### Voctiv platform + `key_storage` (TTS credentials)
 *
 * Same idea as ASR: when Voctiv platform compatibility is on and keys are loaded from PostgreSQL, **`authentication_data`**
 * contains **`legacyTtsKeysByName`**, keyed by **`key_storage.name`** for **this dialog’s agent
 * and company** (agent is implied by the dialog — no UUID in the script).
 *
 * - **`name`** on **`PlayOptions`**: select that row’s flat credentials for this synthesis.
 * - Alternative: **`ttsConfig.name`** (the **`name`** entry is stripped before vendor params).
 * - If omitted, **`defaultTtsName`** on the channel (Omni / route) applies.
 * - Otherwise credentials come from **`authentication_data.tts.&lt;engine&gt;`** for the resolved vendor.
 *
 * ### Note on the field **`name`**
 *
 * This **`name`** is **not** the same as {@link import('./media-channel').LlmOptions.name}
 * (speaker label for LLM). It only selects the TTS key row when LE key catalogs (`legacyTtsKeysByName`, etc.) are present.
 */
export interface PlayOptions {
    /**
     * Mixer queue index **0–4**. Use separate queues to layer music, earcons, and agent TTS
     * so **`stop(queue)`** does not cut unrelated audio.
     *
     * SIP/WS runtimes pass this to the mixer; invalid indices are not part of the public contract.
     * @defaultValue 0
     */
    queue?: number;
    /**
     * Stable id for this queue item — used with **`remove`**, debug UIs, and completion tracking.
     * Should be unique per logical utterance if you need to cancel a specific playback.
     *
     * For sentence-split TTS the runtime creates item aliases by suffixing
     * this value (`alias-0`, `alias-1`, ...). For raw `play()` and direct streaming TTS,
     * the item uses this alias exactly.
     */
    alias?: string;
    /** When `true`, the item restarts after it finishes until **`stop`** / **`remove`**. */
    loop?: boolean;
    /** Milliseconds of silence between loop iterations (if **`loop`**). */
    loopDelayMs?: number;
    /**
     * Linear gain for the whole mixer queue **0.0** (mute) – **1.0** (unity).
     *
     * This is not per-item volume: setting it on one `say()`/`play()` changes the queue
     * volume for subsequent items until it is changed again.
     */
    volume?: number;
    /** Chunking strategy for TTS; see {@link TtsStrategy}. */
    ttsStrategy?: TtsStrategy;
    /**
     * Force a specific TTS vendor for this call, overriding **`channel.params.ttsVendor`**.
     * Ignored for **`play()`** when the source is raw audio (no synthesis).
     * When {@link tts} is set, the handle's resolved vendor/config take precedence.
     */
    ttsVendor?: TtsVendor;
    /**
     * Pre-warmed TTS session from {@link import('./media-channel').MediaChannel.createTts}.
     * Reuses the cached connector / streaming WebSocket so synthesis skips a fresh SSL handshake.
     * When set, the handle's resolved vendor and connector config are used (overrides
     * {@link ttsVendor} / {@link name} / {@link ttsConfig} for credential resolution).
     */
    tts?: TtsHandle;
    /**
     * logic-executor **`key_storage.name`**: use **`authentication_data.legacyTtsKeysByName[name]`**
     * for credentials. Overrides **`defaultTtsName`** on the channel.
     */
    name?: string;
    /**
     * Parameters passed to the TTS connector (voice id, model, `output_format`, nested JSON, …).
     * Primitives are stringified; objects and arrays are JSON-serialized. Merged after channel
     * defaults and catalog credentials; later wins. The key **`name`** is reserved for the
     * storage-row selector and is removed before sending to the vendor.
     */
    ttsConfig?: Record<string, unknown>;
    /**
     * Enable TTS result caching for `say()`.
     *
     * - **`true`** — read/write TTS file cache only (Redis + filesystem + `tts_cache` table or
     *   LE `_cache` phrase in legacy mode).
     * - **`{ phraseName?, flag?, language? }`** — TTS cache **plus** persist into `record_phrase` /
     *   `record_phrase_file` on the Voctiv platform so {@link import('./define-script').PlatformApi.getRecords}
     *   can retrieve the audio later. `phraseName` defaults to `_cache`.
     */
    cache?: true | CacheOptions;
}
/**
 * Options for {@link ChannelAudio.presay}.
 *
 * Caching and persistence to `record_phrase` always happen in legacy mode.
 * Use `cache` only to override the default phrase name / flag / language.
 */
export interface PresayOptions {
    ttsVendor?: TtsVendor;
    /**
     * Pre-warmed TTS session from {@link import('./media-channel').MediaChannel.createTts}.
     * Same reuse semantics as {@link PlayOptions.tts}.
     */
    tts?: TtsHandle;
    name?: string;
    ttsConfig?: Record<string, unknown>;
    ttsStrategy?: TtsStrategy;
    /** Override persist name/flag/language. Defaults to `{ phraseName: '_cache' }`. */
    cache?: CacheOptions;
}
/**
 * Options for {@link ChannelAudio.preload}.
 *
 * In legacy mode the decoded PCM is always persisted to `record_phrase`.
 * Use `cache` only to override the default phrase name / flag / language.
 */
export interface PreloadOptions {
    /** Override persist name/flag/language. Defaults to `{ phraseName: '_cache' }`. */
    cache?: CacheOptions;
}
/**
 * Per-queue mixer control: volume, observability, and manual removal.
 *
 * Obtain via **`channel.audio.queue(index)`**.
 */
export interface MixerQueueControl {
    /** Queue index **0–4** (matches **`PlayOptions.queue`**). */
    readonly index: number;
    /** Current linear volume **0.0–1.0**; assign to change gain for this entire queue. */
    volume: number;
    /** Emits **`alias`** when an item in this queue starts playing. */
    readonly itemStarted$: Observable<string>;
    /** Emits **`alias`** when an item finishes, is removed, or is skipped by queue clear. */
    readonly itemFinished$: Observable<string>;
    /** Emits whenever the queue becomes empty after all pending/current items are gone
     *  and their PCM has been emitted to the mixer output (not merely dequeued). */
    readonly queueEmpty$: Observable<void>;
    /**
     * Remove a single item by **`alias`**.
     * @param alias - Same string passed in **`PlayOptions.alias`** for that item.
     */
    remove(alias: string): void;
    /** Drop all queued and current items on this queue. */
    clear(): void;
}
//# sourceMappingURL=mixer.d.ts.map