@layercode/js-sdk 1.0.13 → 1.0.15

package/dist/layercode-js-sdk.esm.js

@@ -3474,7 +3474,7 @@ class LayercodeClient {
             sessionId: options.sessionId || null,
             authorizeSessionEndpoint: options.authorizeSessionEndpoint,
             metadata: options.metadata || {},
-            vadEnabled: options.vadEnabled || true,
+            vadResumeDelay: options.vadResumeDelay || 500,
             onConnect: options.onConnect || (() => { }),
             onDisconnect: options.onDisconnect || (() => { }),
             onError: options.onError || (() => { }),
@@ -3491,16 +3491,31 @@ class LayercodeClient {
             sampleRate: 16000, // TODO should be set my fetched pipeline config
         });
         this.vad = null;
-        if (this.options.vadEnabled) {
+        this.ws = null;
+        this.status = 'disconnected';
+        this.userAudioAmplitude = 0;
+        this.agentAudioAmplitude = 0;
+        this.sessionId = options.sessionId || null;
+        this.pushToTalkActive = false;
+        this.vadPausedPlayer = false;
+        this.pushToTalkEnabled = false;
+        this.canInterrupt = false;
+        // Bind event handlers
+        this._handleWebSocketMessage = this._handleWebSocketMessage.bind(this);
+        this._handleDataAvailable = this._handleDataAvailable.bind(this);
+    }
+    _initializeVAD() {
+        console.log('initializing VAD', { pushToTalkEnabled: this.pushToTalkEnabled, canInterrupt: this.canInterrupt });
+        if (!this.pushToTalkEnabled && this.canInterrupt) {
             dist.MicVAD.new({
                 stream: this.wavRecorder.getStream() || undefined,
                 model: 'v5',
                 // baseAssetPath: '/', // Use if bundling model locally
                 // onnxWASMBasePath: '/', // Use if bundling model locally
-                positiveSpeechThreshold: 0.4,
-                negativeSpeechThreshold: 0.4,
-                redemptionFrames: 8,
-                minSpeechFrames: 20,
+                positiveSpeechThreshold: 0.3,
+                negativeSpeechThreshold: 0.2,
+                redemptionFrames: 25, // Number of frames of silence before onVADMisfire or onSpeechEnd is called. Effectively a delay before restarting.
+                minSpeechFrames: 15,
                 preSpeechPadFrames: 0,
                 onSpeechStart: () => {
                     // Only pause agent audio if it's currently playing
@@ -3511,23 +3526,35 @@ class LayercodeClient {
                     }
                     else {
                         console.log('onSpeechStart: WavPlayer is not playing, VAD will not pause.');
-                        this.vadPausedPlayer = false;
                     }
                 },
                 onVADMisfire: () => {
-                    // If the speech detected was for less than minSpeechFrames, this is called instead of onSpeechEnd, and we should resume the assistant audio as it was a false interruption
+                    // If the speech detected was for less than minSpeechFrames, this is called instead of onSpeechEnd, and we should resume the assistant audio as it was a false interruption. We include a configurable delay so the assistant isn't too quick to start speaking again.
                     if (this.vadPausedPlayer) {
-                        console.log('onVADMisfire: VAD paused the player, attempting to resume.');
+                        console.log('onSpeechEnd: VAD paused the player, resuming');
                         this.wavPlayer.play();
                         this.vadPausedPlayer = false; // Reset flag
+                        // Option to extend delay in the case where the transcriber takes longer to detect a new turn
+                        // console.log('onVADMisfire: VAD paused the player, resuming in ' + this.options.vadResumeDelay + 'ms');
+                        // // Add configurable delay before resuming playback
+                        // setTimeout(() => {
+                        //   this.wavPlayer.play();
+                        //   this.vadPausedPlayer = false; // Reset flag
+                        // }, this.options.vadResumeDelay);
                     }
                     else {
                         console.log('onVADMisfire: VAD did not pause the player, no action taken to resume.');
                     }
                 },
-                onSpeechEnd: () => {
-                    // We don't take any action here, as the user speech is for more than minSpeechFrames and is very likely to result in a new turn start from the transcriber
-                },
+                // onSpeechEnd: () => {
+                //   if (this.vadPausedPlayer) {
+                //     console.log('onSpeechEnd: VAD paused the player, resuming');
+                //     this.wavPlayer.play();
+                //     this.vadPausedPlayer = false; // Reset flag
+                //   } else {
+                //     console.log('onSpeechEnd: VAD did not pause the player, not resuming.');
+                //   }
+                // },
             })
                 .then((vad) => {
                 this.vad = vad;
@@ -3538,16 +3565,6 @@ class LayercodeClient {
                 console.error('Error initializing VAD:', error);
             });
         }
-        this.ws = null;
-        this.status = 'disconnected';
-        this.userAudioAmplitude = 0;
-        this.agentAudioAmplitude = 0;
-        this.sessionId = options.sessionId || null;
-        this.pushToTalkActive = false;
-        this.vadPausedPlayer = false;
-        // Bind event handlers
-        this._handleWebSocketMessage = this._handleWebSocketMessage.bind(this);
-        this._handleDataAvailable = this._handleDataAvailable.bind(this);
     }
     /**
      * Updates the connection status and triggers the callback
@@ -3607,10 +3624,9 @@ class LayercodeClient {
                     // Sent from the server to this client when a new user turn is detected
                     console.log('received turn.start from server');
                     console.log(message);
-                    // if (message.role === 'user' && !this.pushToTalkActive) {
-                    if (message.role === 'user') {
+                    if (message.role === 'user' && !this.pushToTalkEnabled && this.canInterrupt) {
                         // Interrupt any playing assistant audio if this is a turn trigged by the server (and not push to talk, which will have already called interrupt)
-                        console.log('interrupting assistant audio, as user turn has started and pushToTalkActive is false');
+                        console.log('interrupting assistant audio, as user turn has started and pushToTalkEnabled is false');
                         await this._clientInterruptAssistantReplay();
                     }
                     // if (message.role === 'assistant') {
@@ -3724,6 +3740,19 @@ class LayercodeClient {
             this.ws = new WebSocket(`${this._websocketUrl}?${new URLSearchParams({
                 client_session_key: authorizeSessionResponseBody.client_session_key,
             })}`);
+            const config = authorizeSessionResponseBody.config;
+            console.log('config', config);
+            if (config.transcription.trigger === 'push_to_talk') {
+                this.pushToTalkEnabled = true;
+            }
+            else if (config.transcription.trigger === 'automatic') {
+                this.pushToTalkEnabled = false;
+                this.canInterrupt = config.transcription.can_interrupt;
+            }
+            else {
+                throw new Error(`Unknown trigger: ${config.transcription.trigger}`);
+            }
+            this._initializeVAD();
             // Bind the websocket message callbacks
             this.ws.onmessage = this._handleWebSocketMessage;
             this.ws.onopen = () => {