Upload 54 files

components/LiveAssistant.tsx

@@ -1,3 +1,4 @@
+
 import React, { useState, useRef, useEffect } from 'react';
 import { GoogleGenAI, LiveServerMessage, Modality } from "@google/genai";
 import { Mic, X, MessageCircle, Volume2, Power, Play, Square, Loader2, Bot, ChevronDown, RefreshCw } from 'lucide-react';
@@ -101,6 +102,8 @@ export const LiveAssistant: React.FC = () => {
       
       try {
           setIsInitializing(true);
+          console.log("Starting Live Connection...");
+
           // Setup Audio Context
           // @ts-ignore
           const AudioCtor = window.AudioContext || window.webkitAudioContext;
@@ -113,9 +116,11 @@ export const LiveAssistant: React.FC = () => {
           const stream = await navigator.mediaDevices.getUserMedia({ audio: {
               sampleRate: 16000,
               channelCount: 1,
-              echoCancellation: true
+              echoCancellation: true,
+              noiseSuppression: true
           }});
           audioStreamRef.current = stream;
+          console.log("Microphone access granted");
 
           // Initialize Gemini Client
           const client = new GoogleGenAI({ apiKey });
@@ -125,11 +130,12 @@ export const LiveAssistant: React.FC = () => {
               config: {
                   responseModalities: [Modality.AUDIO],
                   speechConfig: { voiceConfig: { prebuiltVoiceConfig: { voiceName: 'Kore' } } },
-                  systemInstruction: { parts: [{ text: "你是一位乐于助人的校园AI助手。请用简短、自然的中文进行语音对话。" }] },
+                  systemInstruction: { parts: [{ text: "你是一位乐于助人的校园AI助手。请始终使用中文回答。请简短、自然地进行对话，不要使用 Markdown 格式，不要进行搜索。" }] },
                   outputAudioTranscription: { model: true } // Enable transcription to show text
               },
               callbacks: {
                   onopen: () => {
+                      console.log("Live Session Opened");
                       setIsConnected(true);
                       setIsInitializing(false);
                       setLogs(prev => [...prev, {role: 'model', text: '已连接，请点击麦克风说话。'}]);
@@ -165,7 +171,10 @@ export const LiveAssistant: React.FC = () => {
                           // Update last model log or add new
                           setLogs(prev => {
                               const last = prev[prev.length - 1];
-                              if (last && last.role === 'model' && !last.text.endsWith('\n')) {
+                              const isInitialMessage = last && last.text === '已连接，请点击麦克风说话。';
+                              
+                              // IMPORTANT: Do not append to the initial system message
+                              if (last && last.role === 'model' && !isInitialMessage && !last.text.endsWith('\n')) {
                                   // Append to existing turn (simplified logic)
                                   return [...prev.slice(0, -1), { ...last, text: last.text + transcript }];
                               }
@@ -179,6 +188,7 @@ export const LiveAssistant: React.FC = () => {
                       // Note: Standard API usually doesn't echo user transcript in serverContent easily without config, relying on model turn.
                   },
                   onclose: () => {
+                      console.log("Live Session Closed");
                       setIsConnected(false);
                       setLogs(prev => [...prev, {role: 'model', text: '连接已断开'}]);
                   },
@@ -220,6 +230,7 @@ export const LiveAssistant: React.FC = () => {
       
       const newMicState = !isMicOn;
       setIsMicOn(newMicState);
+      console.log("Toggling Mic:", newMicState ? "ON" : "OFF");
 
       if (newMicState) {
           // START SENDING
@@ -233,10 +244,16 @@ export const LiveAssistant: React.FC = () => {
           const source = inputCtx.createMediaStreamSource(audioStreamRef.current);
           const processor = inputCtx.createScriptProcessor(4096, 1, 1);
           
+          let chunkCount = 0;
           processor.onaudioprocess = (e) => {
               if (!newMicState) return; // Guard
               const inputData = e.inputBuffer.getChannelData(0);
               const blob = createBlob(inputData);
+              
+              // Debug log every 20 chunks (~0.5s) to avoid spam but confirm data flow
+              chunkCount++;
+              if (chunkCount % 20 === 0) console.log("Sending audio chunk...", chunkCount);
+
               sessionPromiseRef.current?.then(session => {
                   session.sendRealtimeInput({ media: { mimeType: 'audio/pcm;rate=16000', data: blob.data } });
               });
@@ -254,6 +271,7 @@ export const LiveAssistant: React.FC = () => {
           
       } else {
           // STOP SENDING
+          console.log("Stopping audio stream...");
           if (inputProcessorRef.current) {
               inputProcessorRef.current.disconnect();
               inputProcessorRef.current = null;
@@ -359,4 +377,4 @@ export const LiveAssistant: React.FC = () => {
         )}
     </div>
   );
-};
+};