@layercode/js-sdk 1.0.14 → 1.0.16

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

@@ -3474,7 +3474,6 @@ 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 || (() => { }),
@@ -3492,7 +3491,67 @@ 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;
+        this.userIsSpeaking = 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 we're in push to talk mode, we don't need to use the VAD model
+        if (this.pushToTalkEnabled) {
+            return;
+        }
+        const timeout = setTimeout(() => {
+            console.log('silero vad model timeout');
+            // TODO: send message to server to indicate that the vad model timed out
+            this.userIsSpeaking = true; // allow audio to be sent to the server
+        }, 2000);
+        if (!this.canInterrupt) {
+            dist.MicVAD.new({
+                stream: this.wavRecorder.getStream() || undefined,
+                model: 'v5',
+                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: () => {
+                    if (!this.wavPlayer.isPlaying) {
+                        this.userIsSpeaking = true;
+                    }
+                },
+                onVADMisfire: () => {
+                    this.userIsSpeaking = false;
+                },
+                onSpeechEnd: () => {
+                    this.userIsSpeaking = false;
+                    this._wsSend({
+                        type: 'vad_events',
+                        event: 'vad_end',
+                    });
+                },
+            })
+                .then((vad) => {
+                clearTimeout(timeout);
+                this.vad = vad;
+                this.vad.start();
+                console.log('VAD started');
+            })
+                .catch((error) => {
+                console.error('Error initializing VAD:', error);
+            });
+        }
+        else {
             dist.MicVAD.new({
                 stream: this.wavRecorder.getStream() || undefined,
                 model: 'v5',
@@ -3513,36 +3572,35 @@ class LayercodeClient {
                     else {
                         console.log('onSpeechStart: WavPlayer is not playing, VAD will not pause.');
                     }
+                    this.userIsSpeaking = true;
+                    console.log('onSpeechStart: sending vad_start');
+                    this._wsSend({
+                        type: 'vad_events',
+                        event: 'vad_start',
+                    });
                 },
                 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. We include a configurable delay so the assistant isn't too quick to start speaking again.
+                    this.userIsSpeaking = false;
                     if (this.vadPausedPlayer) {
                         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: () => {
-                //   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.');
-                //   }
-                // },
+                onSpeechEnd: () => {
+                    this.userIsSpeaking = false;
+                    this._wsSend({
+                        type: 'vad_events',
+                        event: 'vad_end',
+                    });
+                },
             })
                 .then((vad) => {
+                clearTimeout(timeout);
                 this.vad = vad;
                 this.vad.start();
                 console.log('VAD started');
@@ -3551,16 +3609,6 @@ 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
@@ -3620,10 +3668,9 @@ 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') {
@@ -3661,10 +3708,13 @@ class LayercodeClient {
     _handleDataAvailable(data) {
         try {
             const base64 = arrayBufferToBase64(data.mono);
-            this._wsSend({
-                type: 'client.audio',
-                content: base64,
-            });
+            const sendAudio = this.pushToTalkEnabled ? this.pushToTalkActive : this.userIsSpeaking;
+            if (sendAudio) {
+                this._wsSend({
+                    type: 'client.audio',
+                    content: base64,
+                });
+            }
         }
         catch (error) {
             console.error('Error processing audio:', error);
@@ -3737,6 +3787,19 @@ 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 = () => {