import type { STT as STTBase } from '../agents/providers.js';
/** Options for {@link STT.deepgram} (Deepgram). */
export type STTDeepgramOptions = {
    /** Provider identifier; normally left unset. */
    provider?: string | null;
    /** API key for the provider. */
    apiKey?: string | null;
    /** Model name. Defaults to `"nova-2"`. */
    modelId?: string | null;
    /** Recognition language (BCP-47, e.g. `"en-US"`). Defaults to `"en-US"`. */
    language?: string | null;
    /** Input audio sample rate in Hz. Defaults to 48000. */
    sampleRate?: number | null;
    /** Silence duration in ms used to detect speech endpoints. */
    endpointing?: number | null;
    /** Emit interim (non-final) transcripts. Defaults to true. */
    interimResults?: boolean | null;
    /** Add punctuation to transcripts. Defaults to true. */
    punctuate?: boolean | null;
    /** Apply smart formatting (dates, numbers, etc.). Defaults to true. */
    smartFormat?: boolean | null;
    /** Include filler words (um, uh) in transcripts. */
    fillerWords?: boolean | null;
    /** Keywords to boost during recognition. */
    keywords?: string[] | null;
    /** Key terms to bias recognition toward. */
    keyterm?: string[] | null;
    /** Override the inference endpoint base URL. */
    baseUrl?: string | null;
};
/** Options for {@link STT.google} (Google). */
export type STTGoogleOptions = {
    /** Provider identifier; normally left unset. */
    provider?: string | null;
    /** API key for the provider. */
    apiKey?: string | null;
    /** Model name. Defaults to `"chirp_3"`. */
    modelId?: string | null;
    /** One or more recognition languages. Defaults to `[language]`. */
    languages?: string | string[] | null;
    /** Single recognition language (BCP-47). Defaults to `"en-US"`. */
    language?: string | null;
    /** Provider region/location. Defaults to `"asia-south1"`. */
    location?: string | null;
    /** Input audio sample rate in Hz. Defaults to 48000. */
    sampleRate?: number | null;
    /** Add punctuation to transcripts. Defaults to true. */
    punctuate?: boolean | null;
    /** Mask profanity in transcripts. */
    profanityFilter?: boolean | null;
    /** Override the inference endpoint base URL. */
    baseUrl?: string | null;
};
/** Options for {@link STT.sarvam} (Sarvam AI). */
export type STTSarvamOptions = {
    /** Provider identifier; normally left unset. */
    provider?: string | null;
    /** API key for the provider. */
    apiKey?: string | null;
    /** Model name. Defaults to `"saaras:v3"`. */
    modelId?: string | null;
    /** Recognition language (BCP-47). Defaults to `"en-IN"`. */
    language?: string | null;
    /** Override the inference endpoint base URL. */
    baseUrl?: string | null;
};
/** Options for {@link STT.assemblyai} (AssemblyAI). */
export type STTAssemblyAIOptions = {
    /** Provider identifier; normally left unset. */
    provider?: string | null;
    /** API key for the provider. */
    apiKey?: string | null;
    /** Provider region. Defaults to `"US"`. */
    region?: string | null;
    /** Input audio sample rate in Hz. Defaults to 48000. */
    inputSampleRate?: number | null;
    /** Sample rate in Hz the audio is resampled to before transcription. Defaults to 16000. */
    targetSampleRate?: number | null;
    /** Format finalized turns (punctuation/casing). Defaults to true. */
    formatTurns?: boolean | null;
    /** Key terms to bias recognition toward. Defaults to none. */
    keytermsPrompt?: string[] | null;
    /** Confidence threshold for detecting the end of a turn, in `[0, 1]`. Defaults to 0.5. */
    endOfTurnConfidenceThreshold?: number | null;
    /** Minimum trailing silence in ms to end a turn when confidence is high. Defaults to 800. */
    minEndOfTurnSilenceWhenConfident?: number | null;
    /** Maximum silence in ms before a turn is forcibly ended. Defaults to 2000. */
    maxTurnSilence?: number | null;
    /** Speech model to use. Defaults to `"universal-streaming-english"`. */
    speechModel?: string | null;
    /** Enable automatic language detection. Defaults to true. */
    languageDetection?: boolean | null;
    /** Additional inference configuration merged last, overriding individual fields. */
    config?: Record<string, any> | null;
    /** Override the inference endpoint base URL. */
    baseUrl?: string | null;
};
/** Options for {@link STT.cartesia} (Cartesia). */
export type STTCartesiaOptions = {
    /** Provider identifier; normally left unset. */
    provider?: string | null;
    /** API key for the provider. */
    apiKey?: string | null;
    /** Model name. Defaults to `"ink-2"`. */
    modelId?: string | null;
    /** Recognition language. Defaults to `"en"`. */
    language?: string | null;
    /** Input audio sample rate in Hz. Defaults to 48000. */
    sampleRate?: number | null;
    /** Override the inference endpoint base URL. */
    baseUrl?: string | null;
};
/** Factory for speech-to-text components used by the direct inference API. */
export declare class STT {
    /** Build a Deepgram speech-to-text component. Defaults: model `"nova-2"`, language `"en-US"`, sample rate 48000 Hz. */
    static deepgram(opts?: STTDeepgramOptions): STTBase;
    /** Build a Google speech-to-text component. Defaults: model `"chirp_3"`, language `"en-US"`, location `"us-central1"`, sample rate 48000 Hz. */
    static google(opts?: STTGoogleOptions): STTBase;
    /** Build a Sarvam AI speech-to-text component. Defaults: model `"saaras:v3"`, language `"en-IN"`, sample rate 48000 Hz. */
    static sarvam(opts?: STTSarvamOptions): STTBase;
    /** Build a Cartesia speech-to-text component. Defaults: model `"ink-2"`, language `"en"`, sample rate 48000 Hz. */
    static cartesia(opts?: STTCartesiaOptions): STTBase;
    /** Build an AssemblyAI speech-to-text component. Defaults: model `"universal-streaming-english"`, language `"en-US"`, region `"US"`, input sample rate 48000 Hz, target sample rate 16000 Hz, format turns on, end-of-turn confidence 0.5, min end-of-turn silence 800 ms, max turn silence 2000 ms, language detection on. */
    static assemblyai(opts?: STTAssemblyAIOptions): STTBase;
}