@layercode/js-sdk 2.8.0 → 2.8.2

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

@@ -5951,6 +5951,7 @@ registerProcessor('audio_processor', AudioProcessor);
 	        this._emitAudioInput();
 	        this.AMPLITUDE_MONITORING_SAMPLE_RATE = 2;
 	        this._websocketUrl = DEFAULT_WS_URL;
+	        this.audioOutputReady = null;
 	        this.wavRecorder = new WavRecorder({ sampleRate: 8000 }); // TODO should be set my fetched agent config
 	        this.wavPlayer = new WavStreamPlayer({
 	            finishedPlayingCallback: this._clientResponseAudioReplayFinished.bind(this),
@@ -5967,6 +5968,7 @@ registerProcessor('audio_processor', AudioProcessor);
 	        this.canInterrupt = false;
 	        this.userIsSpeaking = false;
 	        this.agentIsSpeaking = false;
+	        this.agentIsPlayingAudio = false;
 	        this.recorderStarted = false;
 	        this.readySent = false;
 	        this.currentTurnId = null;
@@ -6112,12 +6114,21 @@ registerProcessor('audio_processor', AudioProcessor);
 	        this.status = status;
 	        this.options.onStatusChange(status);
 	    }
+	    async _waitForAudioOutputReady() {
+	        if (!this.audioOutputReady) {
+	            return;
+	        }
+	        await this.audioOutputReady;
+	    }
 	    _setAgentSpeaking(isSpeaking) {
-	        if (this.agentIsSpeaking === isSpeaking) {
+	        // Track the actual audio playback state regardless of audioOutput setting
+	        this.agentIsPlayingAudio = isSpeaking;
+	        const shouldReportSpeaking = this.audioOutput && isSpeaking;
+	        if (this.agentIsSpeaking === shouldReportSpeaking) {
 	            return;
 	        }
-	        this.agentIsSpeaking = isSpeaking;
-	        this.options.onAgentSpeakingChange(isSpeaking);
+	        this.agentIsSpeaking = shouldReportSpeaking;
+	        this.options.onAgentSpeakingChange(shouldReportSpeaking);
 	    }
 	    _setUserSpeaking(isSpeaking) {
 	        const shouldReportSpeaking = this._shouldCaptureUserAudio() && isSpeaking;
@@ -6185,7 +6196,6 @@ registerProcessor('audio_processor', AudioProcessor);
 	                    if (message.role === 'assistant') {
 	                        // Start tracking new agent turn
 	                        console.debug('Agent turn started, will track new turn ID from audio/text');
-	                        this._setAgentSpeaking(true);
 	                        this._setUserSpeaking(false);
 	                    }
 	                    else if (message.role === 'user' && !this.pushToTalkEnabled) {
@@ -6206,10 +6216,24 @@ registerProcessor('audio_processor', AudioProcessor);
 	                    });
 	                    break;
 	                }
-	                case 'response.audio':
-	                    this._setAgentSpeaking(true);
+	                case 'response.audio': {
+	                    await this._waitForAudioOutputReady();
 	                    const audioBuffer = base64ToArrayBuffer(message.content);
-	                    this.wavPlayer.add16BitPCM(audioBuffer, message.turn_id);
+	                    const hasAudioSamples = audioBuffer.byteLength > 0;
+	                    let audioEnqueued = false;
+	                    if (hasAudioSamples) {
+	                        try {
+	                            const playbackBuffer = this.wavPlayer.add16BitPCM(audioBuffer, message.turn_id);
+	                            audioEnqueued = Boolean(playbackBuffer && playbackBuffer.length > 0);
+	                        }
+	                        catch (error) {
+	                            this._setAgentSpeaking(false);
+	                            throw error;
+	                        }
+	                    }
+	                    else {
+	                        console.debug(`Skipping empty audio response for turn ${message.turn_id}`);
+	                    }
 	                    // TODO: once we've added turn_id to the turn.start msgs sent from teh server, we should move this currentTurnId switching logic to the turn.start msg case. We can then remove the currentTurnId setting logic from the response.audio and response.text cases.
 	                    // Set current turn ID from first audio message, or update if different turn
 	                    if (!this.currentTurnId || this.currentTurnId !== message.turn_id) {
@@ -6218,7 +6242,11 @@ registerProcessor('audio_processor', AudioProcessor);
 	                        // Clean up interrupted tracks, keeping only the current turn
 	                        this.wavPlayer.clearInterruptedTracks(this.currentTurnId ? [this.currentTurnId] : []);
 	                    }
+	                    if (audioEnqueued) {
+	                        this._setAgentSpeaking(true);
+	                    }
 	                    break;
+	                }
 	                case 'response.text':
 	                    // Set turn ID from first text message if not set
 	                    if (!this.currentTurnId) {
@@ -6430,12 +6458,26 @@ registerProcessor('audio_processor', AudioProcessor);
 	            this._emitAudioOutput();
 	            if (state) {
 	                this.wavPlayer.unmute();
+	                // Sync agentSpeaking state with actual playback state when enabling audio output
+	                this._syncAgentSpeakingState();
 	            }
 	            else {
 	                this.wavPlayer.mute();
+	                this._setAgentSpeaking(false);
 	            }
 	        }
 	    }
+	    /**
+	     * Syncs the reported agentSpeaking state with the actual audio playback state.
+	     * Called when audioOutput is enabled to ensure proper state synchronization.
+	     */
+	    _syncAgentSpeakingState() {
+	        const shouldReportSpeaking = this.audioOutput && this.agentIsPlayingAudio;
+	        if (this.agentIsSpeaking !== shouldReportSpeaking) {
+	            this.agentIsSpeaking = shouldReportSpeaking;
+	            this.options.onAgentSpeakingChange(shouldReportSpeaking);
+	        }
+	    }
 	    /** Emitters for audio flags */
 	    _emitAudioInput() {
 	        this.options.audioInputChanged(this.audioInput);
@@ -6491,7 +6533,9 @@ registerProcessor('audio_processor', AudioProcessor);
 	            this.setupVadConfig(config);
 	            // Bind the websocket message callbacks
 	            this.bindWebsocketMessageCallbacks(ws, config);
-	            await this.setupAudioOutput();
+	            const audioOutputReady = this.setupAudioOutput();
+	            this.audioOutputReady = audioOutputReady;
+	            await audioOutputReady;
 	        }
 	        catch (error) {
 	            console.error('Error connecting to Layercode agent:', error);