uneeq-js 3.10.4 → 3.11.0

package/dist/deepgram-flux-stt.d.ts

@@ -0,0 +1,129 @@
+import { type Subject } from 'rxjs';
+import { type UneeqMessage } from './types/UneeqMessages';
+import { type DataChannelMessage } from './webrtc-data-channel/DataChannelMessage';
+import { type PromptMetadata } from './types/PromptMetadata';
+import { type SpeechRecognitionInterface } from './types/SpeechRecognitionInterface';
+export interface DeepgramFluxSTTOptions {
+    connectionUrl: string;
+    jwtToken: string;
+    sessionId: string;
+    model?: string;
+    language?: string;
+    /** End-of-turn confidence threshold (0.5-0.9). @default 0.7 */
+    eotThreshold?: number;
+    /** Eager end-of-turn threshold (0.3-0.9). */
+    eagerEotThreshold?: number;
+    /** End-of-turn timeout in ms. */
+    eotTimeoutMs?: number;
+    /** Send ChatPrompt on EagerEndOfTurn (lower latency, risk of incomplete). @default true */
+    useEagerEndOfTurn?: boolean;
+    /**
+     * Keyterms to boost in transcription results.
+     */
+    keyterms?: string[];
+    echoCancellation?: boolean;
+    noiseSuppression?: boolean;
+    autoGainControl?: boolean;
+    microphoneDeviceId?: string;
+    promptMetadata: PromptMetadata;
+    messages: Subject<UneeqMessage>;
+    sendMessage: (msg: DataChannelMessage) => void;
+}
+export declare class DeepgramFluxSTT implements SpeechRecognitionInterface {
+    private readonly options;
+    private connection;
+    private state;
+    private shouldReconnect;
+    private stream;
+    private audioContext;
+    private workletNode;
+    private reconnectAttempts;
+    private reconnectDelay;
+    private reconnectTimeoutId;
+    private digitalHumanSpeaking;
+    private isUserCurrentlySpeaking;
+    private isUiShowingSpeaking;
+    private eagerPromptSentForTurn;
+    private safetyNetTimeoutId;
+    private audioChunksSent;
+    constructor(options: DeepgramFluxSTTOptions);
+    startRecognition(): Promise<void>;
+    stopRecognition(): Promise<void>;
+    pause(): Promise<boolean>;
+    resume(): Promise<boolean>;
+    setChatMetadata(chatMetadata: PromptMetadata): void;
+    private connect;
+    private disconnect;
+    private scheduleReconnect;
+    private resetReconnectionState;
+    private clearReconnectTimeout;
+    private getToken;
+    private startMicrophone;
+    private stopMicrophone;
+    private setupEventHandlers;
+    /**
+     * Primary handler for Flux v2 TurnInfo events.
+     * Routes to specific handlers based on the turn event type.
+     */
+    private handleTurnInfo;
+    /**
+     * StartOfTurn: User has begun speaking.
+     * Emit UserStartedSpeaking and send UserSpeaking(Start) to data channel.
+     */
+    private handleStartOfTurn;
+    /**
+     * Update: Interim transcript during the current turn.
+     *
+     * This is also the interruption trigger: the first Update with real transcript
+     * while the avatar is speaking sends StopSpeaking immediately. No word thresholds —
+     * Flux's own turn detection is the source of truth.
+     */
+    private handleUpdate;
+    /**
+     * EagerEndOfTurn: Model predicts the user might be done speaking.
+     * If useEagerEndOfTurn is enabled, send ChatPrompt immediately for lower latency.
+     *
+     * If the avatar is still speaking (interruption disabled or Update hasn't fired yet),
+     * the prompt is discarded — we can't send competing prompts while the avatar talks.
+     */
+    private handleEagerEndOfTurn;
+    /**
+     * TurnResumed: Model detects the user continued speaking after an EagerEndOfTurn.
+     * If an eager prompt was already sent, log a warning (can't unsend).
+     */
+    private handleTurnResumed;
+    /**
+     * EndOfTurn: Model has determined the user has finished speaking.
+     *
+     * If the avatar is still speaking (interruption disabled), the prompt is discarded.
+     * Otherwise, cancel any in-flight eager prompt and send the definitive transcript.
+     */
+    private handleEndOfTurn;
+    /**
+     * Handle FatalError from v2 API.
+     */
+    private handleFatalError;
+    /**
+     * Calculate average confidence from word-level data, or return a default.
+     */
+    private calculateWordConfidence;
+    private handleAppMessages;
+    private handleConnectionOpen;
+    private handleConnectionClose;
+    private handleError;
+    private sendChatPrompt;
+    private handleSpeakingEnd;
+    /**
+     * Reset the safety net timer. Called on every TurnInfo event.
+     * If the user is in a speaking state and no events arrive for SAFETY_NET_TIMEOUT_MS,
+     * the speaking indicators are reset to prevent the UI getting stuck.
+     */
+    private resetSafetyNet;
+    private clearSafetyNet;
+    /**
+     * Reset speaking states and send appropriate stop messages.
+     */
+    private resetSpeakingStates;
+    private dataChannelMsgSend;
+    private clientMsgSend;
+}

package/dist/deepgram-flux-stt.spec.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/index.d.ts

@@ -6,5 +6,6 @@ export type { SpeechTranscriptionResult } from './types/SpeechTranscriptionResul
 export { CameraAnchorDistance, CameraAnchorHorizontal } from './webrtc-data-channel/messages/CameraAnchor';
 export { LogLevel } from './types/LogLevel';
 export type { DeepgramSTTOptions } from './deepgram-stt';
+export type { DeepgramFluxSTTOptions } from './deepgram-flux-stt';
 export type { NetworkQualityMetrics, NetworkIndicatorConfig } from './types/NetworkQuality';
 export { NetworkQualityLevel, NetworkIndicatorPosition, NetworkIndicatorVisibility, NetworkIndicatorPopupTrigger } from './types/NetworkQuality';