@volley/recognition-client-sdk 0.1.707 → 0.1.782

package/dist/recog-client-sdk.browser.js

@@ -3747,6 +3747,9 @@ var RecognitionProvider;
   RecognitionProvider2["CARTESIA"] = "cartesia";
   RecognitionProvider2["DASHSCOPE"] = "dashscope";
   RecognitionProvider2["BEDROCK"] = "bedrock";
+  RecognitionProvider2["INWORLD_STT"] = "inworld-stt";
+  RecognitionProvider2["AWS_TRANSCRIBE"] = "aws-transcribe";
+  RecognitionProvider2["AMAZON_NOVA_SONIC"] = "amazon-nova-sonic";
   RecognitionProvider2["TEST_ASR_PROVIDER_QUOTA"] = "test-asr-provider-quota";
   RecognitionProvider2["TEST_ASR_STREAMING"] = "test-asr-streaming";
 })(RecognitionProvider || (RecognitionProvider = {}));
@@ -3797,6 +3800,7 @@ var ElevenLabsModel;
 })(ElevenLabsModel || (ElevenLabsModel = {}));
 var OpenAIRealtimeModel;
 (function(OpenAIRealtimeModel2) {
+  OpenAIRealtimeModel2["GPT_REALTIME_WHISPER"] = "gpt-realtime-whisper";
   OpenAIRealtimeModel2["GPT_4O_TRANSCRIBE"] = "gpt-4o-transcribe";
   OpenAIRealtimeModel2["GPT_4O_MINI_TRANSCRIBE"] = "gpt-4o-mini-transcribe";
 })(OpenAIRealtimeModel || (OpenAIRealtimeModel = {}));
@@ -3819,6 +3823,19 @@ var BedrockModel;
   BedrockModel2["VOXTRAL_MINI_3B_2507"] = "mistral.voxtral-mini-3b-2507";
   BedrockModel2["VOXTRAL_SMALL_24B_2507"] = "mistral.voxtral-small-24b-2507";
 })(BedrockModel || (BedrockModel = {}));
+var InworldSttModel;
+(function(InworldSttModel2) {
+  InworldSttModel2["INWORLD_STT_1"] = "inworld/inworld-stt-1";
+})(InworldSttModel || (InworldSttModel = {}));
+var AwsTranscribeModel;
+(function(AwsTranscribeModel2) {
+  AwsTranscribeModel2["DEFAULT"] = "default";
+})(AwsTranscribeModel || (AwsTranscribeModel = {}));
+var AmazonNovaSonicModel;
+(function(AmazonNovaSonicModel2) {
+  AmazonNovaSonicModel2["AMAZON_NOVA_SONIC_V1"] = "amazon.nova-sonic-v1:0";
+  AmazonNovaSonicModel2["AMAZON_NOVA_2_SONIC"] = "amazon.nova-2-sonic-v1:0";
+})(AmazonNovaSonicModel || (AmazonNovaSonicModel = {}));
 var SelfServeVllmModel;
 (function(SelfServeVllmModel2) {
   SelfServeVllmModel2["QWEN3_ASR_1_7B"] = "qwen3-asr-1.7b";
@@ -3835,6 +3852,18 @@ var RecognitionResultTypeV1;
   RecognitionResultTypeV12["AUDIO_METRICS"] = "AudioMetrics";
   RecognitionResultTypeV12["SESSION_CONFIGURED"] = "SessionConfigured";
 })(RecognitionResultTypeV1 || (RecognitionResultTypeV1 = {}));
+var DetectionTypeV1;
+(function(DetectionTypeV12) {
+  DetectionTypeV12["SEARCH"] = "search";
+})(DetectionTypeV1 || (DetectionTypeV1 = {}));
+var DetectionV1Schema = z.object({
+  type: z.nativeEnum(DetectionTypeV1),
+  query: z.string(),
+  score: z.number().min(0).max(1),
+  startMs: z.number().optional(),
+  endMs: z.number().optional()
+  // Audio time (ms from stream start) where the hit ends
+});
 var TranscriptionResultSchemaV1 = z.object({
   type: z.literal(RecognitionResultTypeV1.TRANSCRIPTION),
   audioUtteranceId: z.string(),
@@ -3853,8 +3882,9 @@ var TranscriptionResultSchemaV1 = z.object({
   endTimestamp: z.number().optional(),
   receivedAtMs: z.number().optional(),
   accumulatedAudioTimeMs: z.number().optional(),
-  rawAudioTimeMs: z.number().optional()
-  // Total audio duration sent to provider (includes prefix)
+  rawAudioTimeMs: z.number().optional(),
+  detections: z.array(DetectionV1Schema).optional()
+  // Provider-reported phrase detections (query + score, optionally startMs/endMs). Always populated when the provider returns hits, regardless of `appendSearch`. Other providers leave this undefined.
 });
 var FunctionCallResultSchemaV1 = z.object({
   type: z.literal(RecognitionResultTypeV1.FUNCTION_CALL),
@@ -3944,9 +3974,9 @@ var ErrorResultSchemaV1 = z.object({
   // Detailed description
 });
 var ClientControlActionV1;
-(function(ClientControlActionV13) {
-  ClientControlActionV13["READY_FOR_UPLOADING_RECORDING"] = "ready_for_uploading_recording";
-  ClientControlActionV13["STOP_RECORDING"] = "stop_recording";
+(function(ClientControlActionV12) {
+  ClientControlActionV12["READY_FOR_UPLOADING_RECORDING"] = "ready_for_uploading_recording";
+  ClientControlActionV12["STOP_RECORDING"] = "stop_recording";
 })(ClientControlActionV1 || (ClientControlActionV1 = {}));
 var ClientControlActionsV1 = z.nativeEnum(ClientControlActionV1);
 var ClientControlMessageSchemaV1 = z.object({
@@ -3979,6 +4009,8 @@ var AudioMetricsResultSchemaV1 = z.object({
   maxVolume: z.number(),
   minVolume: z.number(),
   avgVolume: z.number(),
+  peakVolumeDb: z.number().nullable(),
+  avgVolumeDb: z.number().nullable(),
   silenceRatio: z.number(),
   clippingRatio: z.number(),
   snrEstimate: z.number().nullable(),
@@ -3995,7 +4027,8 @@ var RecognitionResultSchemaV1 = z.discriminatedUnion("type", [
   // P1 - P2
   FunctionCallResultSchemaV1,
   ClientControlMessageSchemaV1,
-  SessionConfiguredSchemaV1
+  SessionConfiguredSchemaV1,
+  AudioMetricsResultSchemaV1
 ]);
 
 // ../../libs/types/dist/provider-transcription.types.js
@@ -4104,7 +4137,15 @@ var TranscriptMessageSchema = z.object({
    * @example true
    * @default false
    */
-  is_fallback: z.boolean().optional()
+  is_fallback: z.boolean().optional(),
+  /**
+   * Provider-reported phrase detections (query + score, optionally
+   * startMs/endMs). Always populated when the provider returns hits,
+   * regardless of `appendSearch` or scene gating. Other providers leave
+   * this undefined.
+   * @example [{ query: 'justin bieber one time', score: 0.78, startMs: 1200, endMs: 2800 }]
+   */
+  detections: z.array(DetectionV1Schema).optional()
 });
 var VADEndSignalSchema = z.object({
   type: z.literal(ProviderMessageType.VAD_END_SIGNAL),
@@ -4418,6 +4459,12 @@ var ASRRequestSchemaV1 = z.object({
   prefixMode: z.nativeEnum(PrefixMode).optional().default(PrefixMode.NONE),
   prefixId: z.string().optional(),
   prefixTextToRemove: z.array(z.string()).optional(),
+  // Streaming audio metrics opt-in: when > 0, server emits AudioMetrics results throttled to this interval (ms).
+  // Undefined / 0 disables streaming audio metrics (final metrics still embedded in Metadata).
+  audioMetricsIntervalMs: z.number().optional(),
+  // Opt-in: round-trip Deepgram `search` phrase hits into the transcript.
+  // Active only when (model = deepgram nova-2) AND (GameContext.gamePhase = 'Solve Puzzle'). See ASRRequestConfig.appendSearch in asr-config.types.ts for full semantics.
+  appendSearch: z.boolean().optional(),
   // Debug options (FOR DEBUG/TESTING ONLY - not for production use)
   debugCommand: RequestDebugCommandSchema
 });
@@ -5221,6 +5268,7 @@ var MessageHandler = class {
   /**
    * Handle incoming WebSocket message
    */
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
   handleMessage(msg) {
     if (this.callbacks.logger) {
       this.callbacks.logger("debug", "[RecogSDK] Received WebSocket message", {
@@ -5259,6 +5307,9 @@ var MessageHandler = class {
       case RecognitionResultTypeV1.SESSION_CONFIGURED:
         this.callbacks.onSessionConfigured?.(msgData);
         break;
+      case RecognitionResultTypeV1.AUDIO_METRICS:
+        this.callbacks.onAudioMetrics?.(msgData);
+        break;
       default:
         if (this.callbacks.logger) {
           this.callbacks.logger("debug", "[RecogSDK] Unknown message type", { type: msgType });
@@ -5380,6 +5431,7 @@ var RealTimeTwoWayWebSocketRecognitionClient = class _RealTimeTwoWayWebSocketRec
       onMetadata: config.onMetadata || (() => {
       }),
       onSessionConfigured: config.onSessionConfigured,
+      onAudioMetrics: config.onAudioMetrics,
       onError: config.onError || (() => {
       }),
       onConnected: config.onConnected || (() => {
@@ -5408,6 +5460,7 @@ var RealTimeTwoWayWebSocketRecognitionClient = class _RealTimeTwoWayWebSocketRec
       onError: this.config.onError,
       onControlMessage: this.handleControlMessage.bind(this),
       onSessionConfigured: this.config.onSessionConfigured,
+      onAudioMetrics: this.config.onAudioMetrics,
       ...this.config.logger && { logger: this.config.logger }
     });
   }
@@ -5733,6 +5786,16 @@ var RealTimeTwoWayWebSocketRecognitionClient = class _RealTimeTwoWayWebSocketRec
         // Include prefix text to remove if provided (for server-side prefix text removal)
         ...this.config.asrRequestConfig.prefixTextToRemove && {
           prefixTextToRemove: this.config.asrRequestConfig.prefixTextToRemove
+        },
+        // Streaming audio metrics opt-in (ms interval). Server only forwards metrics if > 0.
+        ...this.config.asrRequestConfig.audioMetricsIntervalMs !== void 0 && {
+          audioMetricsIntervalMs: this.config.asrRequestConfig.audioMetricsIntervalMs
+        },
+        // Opt-in: round-trip Deepgram nova-2 search-phrase hits into the
+        // transcript. Only fires server-side when (model = nova-2) AND
+        // (GameContext.gamePhase = 'Solve Puzzle'). See ASRRequestConfig.appendSearch.
+        ...this.config.asrRequestConfig.appendSearch !== void 0 && {
+          appendSearch: this.config.asrRequestConfig.appendSearch
         }
       };
       super.sendMessage(