@glydeunity/voice-sdk 1.2.1 → 1.2.2

package/dist/voice-sdk.es.js

@@ -174,15 +174,15 @@ class g {
    * @returns Voice configuration including system prompt, tools, and Deepgram settings
    */
   async fetchConfig() {
-    const e = `${this.unityUrl}/api/unity/voice/config/${this.config.contextType}`, t = this.config.contextId ? `${e}/${this.config.contextId}` : e, i = await fetch(t, {
+    const e = `${this.unityUrl}/api/unity/voice/config/${this.config.contextType}`, t = this.config.contextId ? `${e}/${this.config.contextId}` : e, s = await fetch(t, {
       method: "GET",
       headers: this.getAuthHeaders()
     });
-    if (!i.ok) {
-      const a = await i.json();
+    if (!s.ok) {
+      const a = await s.json();
       throw new Error(a.error?.message || a.message || "Failed to fetch voice config");
     }
-    const { data: o } = await i.json();
+    const { data: o } = await s.json();
     return o;
   }
   /**
@@ -204,17 +204,48 @@ class g {
           body: JSON.stringify(e)
         });
         if (!t.ok) {
-          const s = await t.json();
-          throw new Error(s.error?.message || s.message || "Failed to authenticate voice session");
+          const i = await t.json();
+          throw new Error(i.error?.message || i.message || "Failed to authenticate voice session");
         }
-        const { data: i } = await t.json(), { token: o, agent_config: a, deepgram_config: r } = i, l = this.config.systemPrompt || a.instructions || this.serverConfig?.system_prompt || "You are a helpful AI assistant.";
+        const { data: s } = await t.json(), { token: o, agent_config: a, deepgram_config: r } = s, l = this.config.systemPrompt || a.instructions || this.serverConfig?.system_prompt || "You are a helpful AI assistant.";
         await this.initializeAudio();
         const c = "wss://agent.deepgram.com/v1/agent/converse";
         this.ws = new WebSocket(c, ["bearer", o]), this.ws.onopen = () => {
-          const s = this.config.deepgramConfig || r || this.serverConfig?.deepgram_config || {
-            think: { provider: { type: "open_ai", model: "gpt-4o-mini" } },
+          const i = this.config.deepgramConfig || r || this.serverConfig?.deepgram_config || {
+            think: {
+              provider: { type: "open_ai", model: "gpt-4.1-mini" },
+              functions: [
+                {
+                  name: "end_conversation",
+                  description: `You are an AI assistant that monitors conversations and ends them when specific stop phrases are detected.
+
+Here is a list of phrases to listen for but not restricted to:
+-stop
+-shut up
+-go away
+-turn off
+-stop listening
+
+Before ending the conversation, always say a brief, polite goodbye such as "Goodbye!", "Take care!", or "Have a great day!".
+
+When monitoring the conversation, pay close attention to any input that matches or closely resembles the phrases listed above. The matching should be case-insensitive and allow for minor variations or typos.
+
+End the conversation immediately if:
+1. The user's input exactly matches any phrase in the list.
+2. The user's input is a close variation of any phrase in the list (e.g., "please shut up" instead of "shut up").
+3. The user's input clearly expresses a desire to end the conversation, even if it doesn't use the exact phrases listed.`,
+                  parameters: {
+                    type: "object",
+                    properties: {
+                      item: { type: "string", description: "The phrase or text that triggered the end of conversation" }
+                    },
+                    required: ["item"]
+                  }
+                }
+              ]
+            },
             speak: { provider: { type: "deepgram", model: "aura-2-thalia-en" } },
-            listen: { provider: { type: "deepgram", model: "nova-2", version: "latest" } }
+            listen: { provider: { type: "deepgram", version: "v2", model: "flux-general-en" } }
           }, h = {
             type: "Settings",
             audio: {
@@ -230,22 +261,38 @@ class g {
             },
             agent: {
               language: "en",
-              speak: s.speak || {
+              speak: i.speak || {
                 provider: { type: "deepgram", model: "aura-2-thalia-en" }
               },
-              listen: s.listen || {
+              listen: i.listen || {
                 provider: { type: "deepgram", version: "v2", model: "flux-general-en" }
               },
               think: {
-                provider: s.think?.provider || { type: "open_ai", model: "gpt-4o-mini" },
-                functions: s.think?.functions || [
+                provider: i.think?.provider || { type: "open_ai", model: "gpt-4.1-mini" },
+                functions: i.think?.functions || [
                   {
                     name: "end_conversation",
-                    description: "End the conversation when stop phrases are detected.",
+                    description: `You are an AI assistant that monitors conversations and ends them when specific stop phrases are detected.
+
+Here is a list of phrases to listen for but not restricted to:
+-stop
+-shut up
+-go away
+-turn off
+-stop listening
+
+Before ending the conversation, always say a brief, polite goodbye such as "Goodbye!", "Take care!", or "Have a great day!".
+
+When monitoring the conversation, pay close attention to any input that matches or closely resembles the phrases listed above. The matching should be case-insensitive and allow for minor variations or typos.
+
+End the conversation immediately if:
+1. The user's input exactly matches any phrase in the list.
+2. The user's input is a close variation of any phrase in the list (e.g., "please shut up" instead of "shut up").
+3. The user's input clearly expresses a desire to end the conversation, even if it doesn't use the exact phrases listed.`,
                     parameters: {
                       type: "object",
                       properties: {
-                        item: { type: "string", description: "The phrase that triggered end of conversation" }
+                        item: { type: "string", description: "The phrase or text that triggered the end of conversation" }
                       },
                       required: ["item"]
                     }
@@ -258,10 +305,10 @@ class g {
           this.ws.send(JSON.stringify(h)), this.emit({ type: "open", payload: { config: a, serverConfig: this.serverConfig } });
         };
         const n = l;
-        this.ws.onmessage = (s) => {
-          if (typeof s.data == "string") {
+        this.ws.onmessage = (i) => {
+          if (typeof i.data == "string") {
             try {
-              if (JSON.parse(s.data).type === "SettingsApplied") {
+              if (JSON.parse(i.data).type === "SettingsApplied") {
                 const p = {
                   type: "UpdatePrompt",
                   prompt: n
@@ -270,10 +317,10 @@ class g {
               }
             } catch {
             }
-            this.handleTextMessage(s.data);
-          } else s.data instanceof Blob ? this.handleAudioData(s.data) : s.data instanceof ArrayBuffer && this.handleAudioBuffer(s.data);
-        }, this.ws.onerror = (s) => {
-          console.error("[GlydeVoice] WebSocket error:", s), this.emit({ type: "error", payload: s });
+            this.handleTextMessage(i.data);
+          } else i.data instanceof Blob ? this.handleAudioData(i.data) : i.data instanceof ArrayBuffer && this.handleAudioBuffer(i.data);
+        }, this.ws.onerror = (i) => {
+          console.error("[GlydeVoice] WebSocket error:", i), this.emit({ type: "error", payload: i });
         }, this.ws.onclose = () => {
           this.cleanup(), this.emit({ type: "close" });
         }, this.renderUI();
@@ -307,8 +354,8 @@ class g {
     } finally {
       URL.revokeObjectURL(e), URL.revokeObjectURL(t);
     }
-    this.playbackWorkletNode = new AudioWorkletNode(this.audioContext, "audio-playback-processor"), this.playbackWorkletNode.connect(this.audioContext.destination), this.playbackWorkletNode.port.onmessage = (i) => {
-      const { type: o } = i.data;
+    this.playbackWorkletNode = new AudioWorkletNode(this.audioContext, "audio-playback-processor"), this.playbackWorkletNode.connect(this.audioContext.destination), this.playbackWorkletNode.port.onmessage = (s) => {
+      const { type: o } = s.data;
       (o === "cleared" || o === "bufferEmpty") && (this.isAgentSpeaking = !1, this.agentAudioDoneReceived = !1, this.emit({ type: "agent_speaking", payload: !1 }));
     };
   }
@@ -332,8 +379,8 @@ class g {
           break;
         case "ConversationText":
           if (t.content && t.content.trim()) {
-            const i = t.role === "assistant" ? "agent" : "user";
-            this.config.onTranscript && this.config.onTranscript(t.content, i), this.emit({ type: "transcript", payload: { text: t.content, role: i } }), this.saveTranscript(t.content, t.role);
+            const s = t.role === "assistant" ? "agent" : "user";
+            this.config.onTranscript && this.config.onTranscript(t.content, s), this.emit({ type: "transcript", payload: { text: t.content, role: s } }), this.saveTranscript(t.content, t.role);
           }
           break;
         case "AgentStartedSpeaking":
@@ -366,9 +413,9 @@ class g {
     this.audioContext.state === "suspended" && this.audioContext.resume();
     const t = e.byteLength;
     if (t === 0) return;
-    const i = t - t % 2;
-    if (i === 0) return;
-    const o = i === t ? e : e.slice(0, i), a = new Int16Array(o), r = new Float32Array(a.length);
+    const s = t - t % 2;
+    if (s === 0) return;
+    const o = s === t ? e : e.slice(0, s), a = new Int16Array(o), r = new Float32Array(a.length);
     for (let n = 0; n < a.length; n++)
       r[n] = a[n] / 32768;
     const l = this.resample24kTo48k(r);
@@ -383,13 +430,13 @@ class g {
    * Resample audio from 24kHz to 48kHz using linear interpolation
    */
   resample24kTo48k(e) {
-    const t = e.length * 2, i = new Float32Array(t);
+    const t = e.length * 2, s = new Float32Array(t);
     for (let a = 0; a < e.length - 1; a++) {
       const r = e[a], l = e[a + 1];
-      i[a * 2] = r, i[a * 2 + 1] = (r + l) / 2;
+      s[a * 2] = r, s[a * 2 + 1] = (r + l) / 2;
     }
     const o = e.length - 1;
-    return i[o * 2] = e[o], i[o * 2 + 1] = e[o], i;
+    return s[o * 2] = e[o], s[o * 2 + 1] = e[o], s;
   }
   /**
    * Clear the playback buffer (for interruption handling)

package/dist/voice-sdk.umd.js

@@ -130,7 +130,39 @@ class AudioPlaybackProcessor extends AudioWorkletProcessor {
 }
 
 registerProcessor('audio-playback-processor', AudioPlaybackProcessor);
-`;class f{config;unityUrl;active=!1;serverConfig=null;ws=null;audioContext=null;mediaStream=null;captureWorkletNode=null;playbackWorkletNode=null;isMuted=!1;outputSampleRate=24e3;inputSampleRate=48e3;isAgentSpeaking=!1;agentAudioDoneReceived=!1;constructor(e){this.config=e,this.unityUrl=e.unityBaseUrl||"https://api.glydeunity.com",!e.publishableKey&&!e.apiKey&&!e.authToken&&console.warn("[GlydeVoice] No authentication method provided. One of publishableKey, apiKey, or authToken is required.")}getAuthHeaders(){const e={"Content-Type":"application/json"};return this.config.publishableKey&&(e["x-publishable-key"]=this.config.publishableKey),this.config.apiKey&&(e["x-api-key"]=this.config.apiKey),this.config.authToken&&(e.Authorization=`Bearer ${this.config.authToken}`),e}async fetchConfig(){const e=`${this.unityUrl}/api/unity/voice/config/${this.config.contextType}`,t=this.config.contextId?`${e}/${this.config.contextId}`:e,i=await fetch(t,{method:"GET",headers:this.getAuthHeaders()});if(!i.ok){const o=await i.json();throw new Error(o.error?.message||o.message||"Failed to fetch voice config")}const{data:a}=await i.json();return a}async start(){if(!this.active){this.active=!0;try{this.config.systemPrompt||(this.serverConfig=await this.fetchConfig(),console.log("[GlydeVoice] Fetched config:",this.serverConfig));const e={context_id:this.config.contextId,domain:typeof window<"u"?window.location.hostname:"localhost"};this.config.systemPrompt&&(e.system_prompt=this.config.systemPrompt),this.config.deepgramConfig&&(e.deepgram_config=this.config.deepgramConfig);const t=await fetch(`${this.unityUrl}/api/unity/voice/auth`,{method:"POST",headers:this.getAuthHeaders(),body:JSON.stringify(e)});if(!t.ok){const s=await t.json();throw new Error(s.error?.message||s.message||"Failed to authenticate voice session")}const{data:i}=await t.json(),{token:a,agent_config:o,deepgram_config:r}=i,h=this.config.systemPrompt||o.instructions||this.serverConfig?.system_prompt||"You are a helpful AI assistant.";await this.initializeAudio();const d="wss://agent.deepgram.com/v1/agent/converse";this.ws=new WebSocket(d,["bearer",a]),this.ws.onopen=()=>{const s=this.config.deepgramConfig||r||this.serverConfig?.deepgram_config||{think:{provider:{type:"open_ai",model:"gpt-4o-mini"}},speak:{provider:{type:"deepgram",model:"aura-2-thalia-en"}},listen:{provider:{type:"deepgram",model:"nova-2",version:"latest"}}},p={type:"Settings",audio:{input:{encoding:"linear16",sample_rate:this.inputSampleRate},output:{encoding:"linear16",sample_rate:this.outputSampleRate,container:"none"}},agent:{language:"en",speak:s.speak||{provider:{type:"deepgram",model:"aura-2-thalia-en"}},listen:s.listen||{provider:{type:"deepgram",version:"v2",model:"flux-general-en"}},think:{provider:s.think?.provider||{type:"open_ai",model:"gpt-4o-mini"},functions:s.think?.functions||[{name:"end_conversation",description:"End the conversation when stop phrases are detected.",parameters:{type:"object",properties:{item:{type:"string",description:"The phrase that triggered end of conversation"}},required:["item"]}}]},greeting:"Hi! I'm ready to speak with you. How can I help you today?"}};this.ws.send(JSON.stringify(p)),this.emit({type:"open",payload:{config:o,serverConfig:this.serverConfig}})};const l=h;this.ws.onmessage=s=>{if(typeof s.data=="string"){try{if(JSON.parse(s.data).type==="SettingsApplied"){const g={type:"UpdatePrompt",prompt:l};this.ws.send(JSON.stringify(g)),this.startMicrophone()}}catch{}this.handleTextMessage(s.data)}else s.data instanceof Blob?this.handleAudioData(s.data):s.data instanceof ArrayBuffer&&this.handleAudioBuffer(s.data)},this.ws.onerror=s=>{console.error("[GlydeVoice] WebSocket error:",s),this.emit({type:"error",payload:s})},this.ws.onclose=()=>{this.cleanup(),this.emit({type:"close"})},this.renderUI()}catch(e){throw console.error("[GlydeVoice] Error starting session:",e),this.active=!1,this.emit({type:"error",payload:e}),e}}}createWorkletBlobUrl(e){const t=new Blob([e],{type:"application/javascript"});return URL.createObjectURL(t)}async initializeAudio(){this.audioContext=new AudioContext({sampleRate:this.inputSampleRate});const e=this.createWorkletBlobUrl(c),t=this.createWorkletBlobUrl(u);try{await Promise.all([this.audioContext.audioWorklet.addModule(e),this.audioContext.audioWorklet.addModule(t)])}finally{URL.revokeObjectURL(e),URL.revokeObjectURL(t)}this.playbackWorkletNode=new AudioWorkletNode(this.audioContext,"audio-playback-processor"),this.playbackWorkletNode.connect(this.audioContext.destination),this.playbackWorkletNode.port.onmessage=i=>{const{type:a}=i.data;(a==="cleared"||a==="bufferEmpty")&&(this.isAgentSpeaking=!1,this.agentAudioDoneReceived=!1,this.emit({type:"agent_speaking",payload:!1}))}}handleTextMessage(e){try{const t=JSON.parse(e);switch(t.type){case"Welcome":this.emit({type:"ready"});break;case"SettingsApplied":break;case"UserStartedSpeaking":this.emit({type:"user_speaking",payload:!0}),this.clearPlaybackBuffer(),this.isAgentSpeaking=!1,this.agentAudioDoneReceived=!1;break;case"UserStoppedSpeaking":this.emit({type:"user_speaking",payload:!1});break;case"ConversationText":if(t.content&&t.content.trim()){const i=t.role==="assistant"?"agent":"user";this.config.onTranscript&&this.config.onTranscript(t.content,i),this.emit({type:"transcript",payload:{text:t.content,role:i}}),this.saveTranscript(t.content,t.role)}break;case"AgentStartedSpeaking":this.isAgentSpeaking=!0,this.agentAudioDoneReceived=!1,this.emit({type:"agent_speaking",payload:!0});break;case"AgentAudioDone":this.agentAudioDoneReceived=!0;break;case"Error":console.error("[GlydeVoice] Agent error:",t),this.emit({type:"error",payload:t});break}}catch(t){console.error("[GlydeVoice] Failed to parse message:",t)}}async handleAudioData(e){const t=await e.arrayBuffer();this.handleAudioBuffer(t)}handleAudioBuffer(e){if(!this.playbackWorkletNode||!this.audioContext)return;this.audioContext.state==="suspended"&&this.audioContext.resume();const t=e.byteLength;if(t===0)return;const i=t-t%2;if(i===0)return;const a=i===t?e:e.slice(0,i),o=new Int16Array(a),r=new Float32Array(o.length);for(let l=0;l<o.length;l++)r[l]=o[l]/32768;const h=this.resample24kTo48k(r);!this.isAgentSpeaking&&!this.agentAudioDoneReceived&&(this.isAgentSpeaking=!0,this.emit({type:"agent_speaking",payload:!0}));const d=new Float32Array(h);this.playbackWorkletNode.port.postMessage({type:"audio",data:d},[d.buffer])}resample24kTo48k(e){const t=e.length*2,i=new Float32Array(t);for(let o=0;o<e.length-1;o++){const r=e[o],h=e[o+1];i[o*2]=r,i[o*2+1]=(r+h)/2}const a=e.length-1;return i[a*2]=e[a],i[a*2+1]=e[a],i}clearPlaybackBuffer(){this.playbackWorkletNode&&this.playbackWorkletNode.port.postMessage({type:"clear"})}async startMicrophone(){if(!this.audioContext)throw new Error("Audio context not initialized");try{this.mediaStream=await navigator.mediaDevices.getUserMedia({audio:{channelCount:1,sampleRate:this.inputSampleRate,echoCancellation:!0,noiseSuppression:!0}});const e=this.audioContext.createMediaStreamSource(this.mediaStream);this.captureWorkletNode=new AudioWorkletNode(this.audioContext,"audio-capture-processor"),this.captureWorkletNode.port.onmessage=t=>{!this.active||!this.ws||this.ws.readyState!==WebSocket.OPEN||this.isMuted||this.ws.send(t.data)},e.connect(this.captureWorkletNode),this.emit({type:"microphone_ready"})}catch(e){throw console.error("[GlydeVoice] Microphone error:",e),e}}async saveTranscript(e,t){if(!(!this.config.contextId||!e))try{await fetch(`${this.unityUrl}/api/unity/voice/transcript`,{method:"POST",headers:this.getAuthHeaders(),body:JSON.stringify({context_id:this.config.contextId,content:e,role:t==="assistant"?"assistant":"user"})})}catch{}}setMuted(e){this.isMuted=e}getMuted(){return this.isMuted}isActive(){return this.active}getServerConfig(){return this.serverConfig}stop(){this.active=!1,this.cleanup()}cleanup(){this.captureWorkletNode&&(this.captureWorkletNode.disconnect(),this.captureWorkletNode.port.close(),this.captureWorkletNode=null),this.playbackWorkletNode&&(this.playbackWorkletNode.disconnect(),this.playbackWorkletNode.port.close(),this.playbackWorkletNode=null),this.mediaStream&&(this.mediaStream.getTracks().forEach(e=>e.stop()),this.mediaStream=null),this.audioContext&&(this.audioContext.close(),this.audioContext=null),this.ws&&(this.ws.readyState===WebSocket.OPEN&&this.ws.close(),this.ws=null)}emit(e){this.config.onEvent&&this.config.onEvent(e)}renderUI(){if(!this.config.container)return;const e=typeof this.config.container=="string"?document.querySelector(this.config.container):this.config.container;e&&(e.innerHTML=`
+`;class f{config;unityUrl;active=!1;serverConfig=null;ws=null;audioContext=null;mediaStream=null;captureWorkletNode=null;playbackWorkletNode=null;isMuted=!1;outputSampleRate=24e3;inputSampleRate=48e3;isAgentSpeaking=!1;agentAudioDoneReceived=!1;constructor(e){this.config=e,this.unityUrl=e.unityBaseUrl||"https://api.glydeunity.com",!e.publishableKey&&!e.apiKey&&!e.authToken&&console.warn("[GlydeVoice] No authentication method provided. One of publishableKey, apiKey, or authToken is required.")}getAuthHeaders(){const e={"Content-Type":"application/json"};return this.config.publishableKey&&(e["x-publishable-key"]=this.config.publishableKey),this.config.apiKey&&(e["x-api-key"]=this.config.apiKey),this.config.authToken&&(e.Authorization=`Bearer ${this.config.authToken}`),e}async fetchConfig(){const e=`${this.unityUrl}/api/unity/voice/config/${this.config.contextType}`,t=this.config.contextId?`${e}/${this.config.contextId}`:e,i=await fetch(t,{method:"GET",headers:this.getAuthHeaders()});if(!i.ok){const o=await i.json();throw new Error(o.error?.message||o.message||"Failed to fetch voice config")}const{data:a}=await i.json();return a}async start(){if(!this.active){this.active=!0;try{this.config.systemPrompt||(this.serverConfig=await this.fetchConfig(),console.log("[GlydeVoice] Fetched config:",this.serverConfig));const e={context_id:this.config.contextId,domain:typeof window<"u"?window.location.hostname:"localhost"};this.config.systemPrompt&&(e.system_prompt=this.config.systemPrompt),this.config.deepgramConfig&&(e.deepgram_config=this.config.deepgramConfig);const t=await fetch(`${this.unityUrl}/api/unity/voice/auth`,{method:"POST",headers:this.getAuthHeaders(),body:JSON.stringify(e)});if(!t.ok){const s=await t.json();throw new Error(s.error?.message||s.message||"Failed to authenticate voice session")}const{data:i}=await t.json(),{token:a,agent_config:o,deepgram_config:r}=i,h=this.config.systemPrompt||o.instructions||this.serverConfig?.system_prompt||"You are a helpful AI assistant.";await this.initializeAudio();const p="wss://agent.deepgram.com/v1/agent/converse";this.ws=new WebSocket(p,["bearer",a]),this.ws.onopen=()=>{const s=this.config.deepgramConfig||r||this.serverConfig?.deepgram_config||{think:{provider:{type:"open_ai",model:"gpt-4.1-mini"},functions:[{name:"end_conversation",description:`You are an AI assistant that monitors conversations and ends them when specific stop phrases are detected.
+
+Here is a list of phrases to listen for but not restricted to:
+-stop
+-shut up
+-go away
+-turn off
+-stop listening
+
+Before ending the conversation, always say a brief, polite goodbye such as "Goodbye!", "Take care!", or "Have a great day!".
+
+When monitoring the conversation, pay close attention to any input that matches or closely resembles the phrases listed above. The matching should be case-insensitive and allow for minor variations or typos.
+
+End the conversation immediately if:
+1. The user's input exactly matches any phrase in the list.
+2. The user's input is a close variation of any phrase in the list (e.g., "please shut up" instead of "shut up").
+3. The user's input clearly expresses a desire to end the conversation, even if it doesn't use the exact phrases listed.`,parameters:{type:"object",properties:{item:{type:"string",description:"The phrase or text that triggered the end of conversation"}},required:["item"]}}]},speak:{provider:{type:"deepgram",model:"aura-2-thalia-en"}},listen:{provider:{type:"deepgram",version:"v2",model:"flux-general-en"}}},d={type:"Settings",audio:{input:{encoding:"linear16",sample_rate:this.inputSampleRate},output:{encoding:"linear16",sample_rate:this.outputSampleRate,container:"none"}},agent:{language:"en",speak:s.speak||{provider:{type:"deepgram",model:"aura-2-thalia-en"}},listen:s.listen||{provider:{type:"deepgram",version:"v2",model:"flux-general-en"}},think:{provider:s.think?.provider||{type:"open_ai",model:"gpt-4.1-mini"},functions:s.think?.functions||[{name:"end_conversation",description:`You are an AI assistant that monitors conversations and ends them when specific stop phrases are detected.
+
+Here is a list of phrases to listen for but not restricted to:
+-stop
+-shut up
+-go away
+-turn off
+-stop listening
+
+Before ending the conversation, always say a brief, polite goodbye such as "Goodbye!", "Take care!", or "Have a great day!".
+
+When monitoring the conversation, pay close attention to any input that matches or closely resembles the phrases listed above. The matching should be case-insensitive and allow for minor variations or typos.
+
+End the conversation immediately if:
+1. The user's input exactly matches any phrase in the list.
+2. The user's input is a close variation of any phrase in the list (e.g., "please shut up" instead of "shut up").
+3. The user's input clearly expresses a desire to end the conversation, even if it doesn't use the exact phrases listed.`,parameters:{type:"object",properties:{item:{type:"string",description:"The phrase or text that triggered the end of conversation"}},required:["item"]}}]},greeting:"Hi! I'm ready to speak with you. How can I help you today?"}};this.ws.send(JSON.stringify(d)),this.emit({type:"open",payload:{config:o,serverConfig:this.serverConfig}})};const l=h;this.ws.onmessage=s=>{if(typeof s.data=="string"){try{if(JSON.parse(s.data).type==="SettingsApplied"){const g={type:"UpdatePrompt",prompt:l};this.ws.send(JSON.stringify(g)),this.startMicrophone()}}catch{}this.handleTextMessage(s.data)}else s.data instanceof Blob?this.handleAudioData(s.data):s.data instanceof ArrayBuffer&&this.handleAudioBuffer(s.data)},this.ws.onerror=s=>{console.error("[GlydeVoice] WebSocket error:",s),this.emit({type:"error",payload:s})},this.ws.onclose=()=>{this.cleanup(),this.emit({type:"close"})},this.renderUI()}catch(e){throw console.error("[GlydeVoice] Error starting session:",e),this.active=!1,this.emit({type:"error",payload:e}),e}}}createWorkletBlobUrl(e){const t=new Blob([e],{type:"application/javascript"});return URL.createObjectURL(t)}async initializeAudio(){this.audioContext=new AudioContext({sampleRate:this.inputSampleRate});const e=this.createWorkletBlobUrl(c),t=this.createWorkletBlobUrl(u);try{await Promise.all([this.audioContext.audioWorklet.addModule(e),this.audioContext.audioWorklet.addModule(t)])}finally{URL.revokeObjectURL(e),URL.revokeObjectURL(t)}this.playbackWorkletNode=new AudioWorkletNode(this.audioContext,"audio-playback-processor"),this.playbackWorkletNode.connect(this.audioContext.destination),this.playbackWorkletNode.port.onmessage=i=>{const{type:a}=i.data;(a==="cleared"||a==="bufferEmpty")&&(this.isAgentSpeaking=!1,this.agentAudioDoneReceived=!1,this.emit({type:"agent_speaking",payload:!1}))}}handleTextMessage(e){try{const t=JSON.parse(e);switch(t.type){case"Welcome":this.emit({type:"ready"});break;case"SettingsApplied":break;case"UserStartedSpeaking":this.emit({type:"user_speaking",payload:!0}),this.clearPlaybackBuffer(),this.isAgentSpeaking=!1,this.agentAudioDoneReceived=!1;break;case"UserStoppedSpeaking":this.emit({type:"user_speaking",payload:!1});break;case"ConversationText":if(t.content&&t.content.trim()){const i=t.role==="assistant"?"agent":"user";this.config.onTranscript&&this.config.onTranscript(t.content,i),this.emit({type:"transcript",payload:{text:t.content,role:i}}),this.saveTranscript(t.content,t.role)}break;case"AgentStartedSpeaking":this.isAgentSpeaking=!0,this.agentAudioDoneReceived=!1,this.emit({type:"agent_speaking",payload:!0});break;case"AgentAudioDone":this.agentAudioDoneReceived=!0;break;case"Error":console.error("[GlydeVoice] Agent error:",t),this.emit({type:"error",payload:t});break}}catch(t){console.error("[GlydeVoice] Failed to parse message:",t)}}async handleAudioData(e){const t=await e.arrayBuffer();this.handleAudioBuffer(t)}handleAudioBuffer(e){if(!this.playbackWorkletNode||!this.audioContext)return;this.audioContext.state==="suspended"&&this.audioContext.resume();const t=e.byteLength;if(t===0)return;const i=t-t%2;if(i===0)return;const a=i===t?e:e.slice(0,i),o=new Int16Array(a),r=new Float32Array(o.length);for(let l=0;l<o.length;l++)r[l]=o[l]/32768;const h=this.resample24kTo48k(r);!this.isAgentSpeaking&&!this.agentAudioDoneReceived&&(this.isAgentSpeaking=!0,this.emit({type:"agent_speaking",payload:!0}));const p=new Float32Array(h);this.playbackWorkletNode.port.postMessage({type:"audio",data:p},[p.buffer])}resample24kTo48k(e){const t=e.length*2,i=new Float32Array(t);for(let o=0;o<e.length-1;o++){const r=e[o],h=e[o+1];i[o*2]=r,i[o*2+1]=(r+h)/2}const a=e.length-1;return i[a*2]=e[a],i[a*2+1]=e[a],i}clearPlaybackBuffer(){this.playbackWorkletNode&&this.playbackWorkletNode.port.postMessage({type:"clear"})}async startMicrophone(){if(!this.audioContext)throw new Error("Audio context not initialized");try{this.mediaStream=await navigator.mediaDevices.getUserMedia({audio:{channelCount:1,sampleRate:this.inputSampleRate,echoCancellation:!0,noiseSuppression:!0}});const e=this.audioContext.createMediaStreamSource(this.mediaStream);this.captureWorkletNode=new AudioWorkletNode(this.audioContext,"audio-capture-processor"),this.captureWorkletNode.port.onmessage=t=>{!this.active||!this.ws||this.ws.readyState!==WebSocket.OPEN||this.isMuted||this.ws.send(t.data)},e.connect(this.captureWorkletNode),this.emit({type:"microphone_ready"})}catch(e){throw console.error("[GlydeVoice] Microphone error:",e),e}}async saveTranscript(e,t){if(!(!this.config.contextId||!e))try{await fetch(`${this.unityUrl}/api/unity/voice/transcript`,{method:"POST",headers:this.getAuthHeaders(),body:JSON.stringify({context_id:this.config.contextId,content:e,role:t==="assistant"?"assistant":"user"})})}catch{}}setMuted(e){this.isMuted=e}getMuted(){return this.isMuted}isActive(){return this.active}getServerConfig(){return this.serverConfig}stop(){this.active=!1,this.cleanup()}cleanup(){this.captureWorkletNode&&(this.captureWorkletNode.disconnect(),this.captureWorkletNode.port.close(),this.captureWorkletNode=null),this.playbackWorkletNode&&(this.playbackWorkletNode.disconnect(),this.playbackWorkletNode.port.close(),this.playbackWorkletNode=null),this.mediaStream&&(this.mediaStream.getTracks().forEach(e=>e.stop()),this.mediaStream=null),this.audioContext&&(this.audioContext.close(),this.audioContext=null),this.ws&&(this.ws.readyState===WebSocket.OPEN&&this.ws.close(),this.ws=null)}emit(e){this.config.onEvent&&this.config.onEvent(e)}renderUI(){if(!this.config.container)return;const e=typeof this.config.container=="string"?document.querySelector(this.config.container):this.config.container;e&&(e.innerHTML=`
         <div style="padding: 20px; border: 1px solid #ccc; border-radius: 8px; background: #fff;">
           <h3>Glyde Voice Agent</h3>
           <p>Status: Active</p>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@glydeunity/voice-sdk",
-  "version": "1.2.1",
+  "version": "1.2.2",
   "description": "GLYDE Voice Agent SDK - AI-powered voice interactions for web applications",
   "type": "module",
   "main": "./dist/voice-sdk.umd.js",