@apteva/apteva-kit 0.1.136 → 0.1.137

package/dist/index.js

@@ -4491,6 +4491,8 @@ function useVoiceSession(config) {
   const mutedRef = _react.useRef.call(void 0, false);
   const configRef = _react.useRef.call(void 0, config);
   configRef.current = config;
+  const agentSpeakingRef = _react.useRef.call(void 0, false);
+  const agentSpeakingTimeoutRef = _react.useRef.call(void 0, null);
   const cleanup = _react.useCallback.call(void 0, () => {
     if (durationIntervalRef.current) {
       clearInterval(durationIntervalRef.current);
@@ -4527,6 +4529,11 @@ function useVoiceSession(config) {
     }
     nextPlayTimeRef.current = 0;
     mutedRef.current = false;
+    agentSpeakingRef.current = false;
+    if (agentSpeakingTimeoutRef.current) {
+      clearTimeout(agentSpeakingTimeoutRef.current);
+      agentSpeakingTimeoutRef.current = null;
+    }
     setMuted(false);
     setPartialTranscript("");
     setDuration(0);
@@ -4554,6 +4561,14 @@ function useVoiceSession(config) {
     const startTime = Math.max(currentTime, nextPlayTimeRef.current);
     source.start(startTime);
     nextPlayTimeRef.current = startTime + audioBuffer.duration;
+    agentSpeakingRef.current = true;
+    if (agentSpeakingTimeoutRef.current) {
+      clearTimeout(agentSpeakingTimeoutRef.current);
+    }
+    const remainingMs = (nextPlayTimeRef.current - currentTime) * 1e3 + 150;
+    agentSpeakingTimeoutRef.current = setTimeout(() => {
+      agentSpeakingRef.current = false;
+    }, remainingMs);
   }, []);
   const startCaptureRef = _react.useRef.call(void 0, () => {
   });
@@ -4617,13 +4632,23 @@ function useVoiceSession(config) {
     if (!ws) return;
     try {
       captureCtxRef.current = new AudioContext();
+      if (captureCtxRef.current.state === "suspended") {
+        await captureCtxRef.current.resume();
+      }
       const nativeSampleRate = captureCtxRef.current.sampleRate;
-      mediaStreamRef.current = await navigator.mediaDevices.getUserMedia({ audio: true });
+      mediaStreamRef.current = await navigator.mediaDevices.getUserMedia({
+        audio: {
+          echoCancellation: true,
+          noiseSuppression: true,
+          autoGainControl: true
+        }
+      });
       const source = captureCtxRef.current.createMediaStreamSource(mediaStreamRef.current);
       processorRef.current = captureCtxRef.current.createScriptProcessor(2048, 1, 1);
       processorRef.current.onaudioprocess = (e) => {
         if (!ws || ws.readyState !== WebSocket.OPEN) return;
         if (mutedRef.current) return;
+        if (agentSpeakingRef.current) return;
         const inputData = e.inputBuffer.getChannelData(0);
         const resampledData = resampleAudio(inputData, nativeSampleRate, 16e3);
         const int16Data = float32ToInt16(resampledData);
@@ -4656,6 +4681,12 @@ function useVoiceSession(config) {
     }
     const ws = new WebSocket(wsUrl);
     wsRef.current = ws;
+    if (!playbackCtxRef.current) {
+      playbackCtxRef.current = new AudioContext({ sampleRate: 24e3 });
+    }
+    if (playbackCtxRef.current.state === "suspended") {
+      playbackCtxRef.current.resume();
+    }
     ws.onopen = () => {
       const provider = configRef.current.provider || "openai";
       const voice = configRef.current.voice || "ash";