修改音频阈值

service/llm_service.go

@@ -616,44 +616,93 @@ func (s *LLMService) TrimAudioSilence(audioData string) (string, error) {
 	channels := int(fmtChunk.NumChannels)
 	bytesPerSample := int(fmtChunk.BitsPerSample) / 8
 
-	// Find the last non-silent sample
+	// 优化后的静音检测参数
+	silenceThreshold := 0.005                                     // 降低静音阈值，更敏感地检测声音
+	windowSize := int(float64(fmtChunk.SampleRate) * 0.05)        // 50ms滑动窗口
+	minSilenceDuration := int(float64(fmtChunk.SampleRate) * 0.1) // 100ms最小静音持续时间
+	bufferSamples := int(float64(fmtChunk.SampleRate) * 0.05)     // 减少缓冲区到50ms
+
+	// 计算RMS能量的辅助函数
+	calculateRMS := func(startSample, endSample int) float64 {
+		if startSample >= endSample || startSample >= samplesPerChannel {
+			return 0.0
+		}
+		if endSample > samplesPerChannel {
+			endSample = samplesPerChannel
+		}
+
+		var sum float64
+		sampleCount := 0
+
+		for i := startSample; i < endSample; i++ {
+			for ch := 0; ch < channels; ch++ {
+				offset := i*int(fmtChunk.BlockAlign) + ch*bytesPerSample
+				if offset+bytesPerSample > len(audioBytes) {
+					continue
+				}
+
+				// Convert bytes to sample value
+				var sample int16
+				if err := binary.Read(bytes.NewReader(audioBytes[offset:offset+bytesPerSample]), binary.LittleEndian, &sample); err != nil {
+					continue
+				}
+
+				// Normalize sample to [-1, 1] range and square it
+				normalizedSample := float64(sample) / 32768.0
+				sum += normalizedSample * normalizedSample
+				sampleCount++
+			}
+		}
+
+		if sampleCount == 0 {
+			return 0.0
+		}
+		return math.Sqrt(sum / float64(sampleCount))
+	}
+
+	// 使用滑动窗口检测静音
 	lastNonSilent := 0
-	silenceThreshold := 0.01 // Adjust this threshold as needed
+	consecutiveSilentWindows := 0
+	requiredSilentWindows := minSilenceDuration / windowSize
 
-	for i := 0; i < samplesPerChannel; i++ {
-		isSilent := true
-		for ch := 0; ch < channels; ch++ {
-			offset := i*int(fmtChunk.BlockAlign) + ch*bytesPerSample
-			if offset+bytesPerSample > len(audioBytes) {
-				continue
-			}
+	for i := 0; i < samplesPerChannel; i += windowSize {
+		endSample := i + windowSize
+		if endSample > samplesPerChannel {
+			endSample = samplesPerChannel
+		}
 
-			// Convert bytes to sample value
-			var sample int16
-			if err := binary.Read(bytes.NewReader(audioBytes[offset:offset+bytesPerSample]), binary.LittleEndian, &sample); err != nil {
-				continue
-			}
+		rms := calculateRMS(i, endSample)
 
-			// Normalize sample to [-1, 1] range
-			normalizedSample := float64(sample) / 32768.0
-			if math.Abs(normalizedSample) > silenceThreshold {
-				isSilent = false
+		if rms > silenceThreshold {
+			// 检测到声音
+			lastNonSilent = endSample
+			consecutiveSilentWindows = 0
+		} else {
+			// 检测到静音
+			consecutiveSilentWindows++
+
+			// 如果连续静音窗口足够多，认为找到了真正的静音开始点
+			if consecutiveSilentWindows >= requiredSilentWindows {
+				lastNonSilent = i
 				break
 			}
 		}
-
-		if !isSilent {
-			lastNonSilent = i
-		}
 	}
 
-	// Add a small buffer (e.g., 0.1 seconds) after the last non-silent sample
-	bufferSamples := int(float64(fmtChunk.SampleRate) * 0.1)
+	// 添加缓冲区，但减少缓冲区大小
 	lastSample := lastNonSilent + bufferSamples
 	if lastSample > samplesPerChannel {
 		lastSample = samplesPerChannel
 	}
 
+	// 确保至少保留一些音频数据
+	if lastSample < int(fmtChunk.SampleRate)*2 { // 至少保留2秒
+		lastSample = int(fmtChunk.SampleRate) * 2
+		if lastSample > samplesPerChannel {
+			lastSample = samplesPerChannel
+		}
+	}
+
 	// Calculate new data size
 	newDataSize := lastSample * int(fmtChunk.BlockAlign)
 	trimmedAudio := audioBytes[:newDataSize]