优化后的代码

app/src/main/java/com/zs/smarthuman/sherpa/VoiceController.kt

@@ -9,7 +9,7 @@ class VoiceController(
     assetManager: AssetManager,
     private val onWakeup: () -> Unit,
     private val onFinalAudio: (FloatArray) -> Unit,
-    private val idleTimeoutSeconds: Int = 8,
+    private val idleTimeoutSeconds: Int = 10,
     private val maxRecordingSeconds: Int = 10,
     private val onStateChanged: ((VoiceState) -> Unit)? = null,
     private val stopBackendAudio: (() -> Unit)? = null
@@ -62,7 +62,6 @@ class VoiceController(
     private val MIN_SPEECH_MS = 1000L
     private val MIN_AVG_ENERGY = 0.02f
 
-    private val WAIT_SPEECH_TIMEOUT_MS = 8000L
 
     /* ================= 音频入口 ================= */
 
@@ -93,7 +92,7 @@ class VoiceController(
 
             VoiceState.WAIT_SPEECH -> {
 
-                if (waitSpeechStartMs > 0 && now - waitSpeechStartMs >= WAIT_SPEECH_TIMEOUT_MS) {
+                if (waitSpeechStartMs > 0 && now - waitSpeechStartMs >= idleTimeoutSeconds * 1000) {
                     Log.d(TAG, "⏱ Wakeup but no speech → WAIT_WAKEUP")
                     resetAll()
                     return
@@ -196,9 +195,17 @@ class VoiceController(
             return
         }
 
+        val audio = audioBuffer.toFloatArray()
         val vadRatio = vadManager.activeSpeechRatio()
         val avgEnergy = if (speechFrameCount > 0) speechEnergySum / speechFrameCount else 0f
 
+        // ⭐ 多人说话检测：短时峰值方差
+        if (shortTermPeakVariance(audio) > 0.015f) {
+            Log.d(TAG, "❌ Short-term peak variance too high → likely multi-speaker")
+            resetToWaitSpeech()
+            return
+        }
+
         var score = 0
         when {
             duration >= 4000 -> score += 3
@@ -230,12 +237,36 @@ class VoiceController(
         }
 
         waitSpeechFailStartMs = 0L
-        val finalAudio = audioBuffer.toFloatArray()
         audioBuffer.clear()
         state = VoiceState.UPLOADING
-        onFinalAudio(finalAudio)
+        onFinalAudio(audio)
     }
 
+    /** 计算短时峰值方差，用于多人说话检测 */
+    private fun shortTermPeakVariance(audio: FloatArray): Float {
+        val frameSize = 160 // 10ms @16kHz
+        val peaks = mutableListOf<Float>()
+        var i = 0
+        while (i + frameSize <= audio.size) {
+            val frame = audio.sliceArray(i until i + frameSize)
+            peaks.add(calcPeakRms(frame))
+            i += frameSize
+        }
+        val mean = peaks.average().toFloat()
+        return peaks.map { (it - mean) * (it - mean) }.average().toFloat()
+    }
+
+    /** 计算音频帧峰值 */
+    private fun calcPeakRms(audio: FloatArray): Float {
+        var peak = 0f
+        for (v in audio) {
+            val abs = kotlin.math.abs(v)
+            if (abs > peak) peak = abs
+        }
+        return peak
+    }
+
+
     /* ================= 播放回调 ================= */
 
     fun onPlayStartPrompt() { state = VoiceState.PLAYING_PROMPT }        // ⭐ 补全