Fix bug where disabling PTT would incorrectly stop VAD capture

cmd/tui/model.go

@@ -97,12 +97,15 @@ type Model struct {
 	talkingClients map[uint16]bool // ClientID -> isTalking
 
 	// Audio state
-	audioPlayer   *audio.Player
+	audioPlayer      *audio.Player
-	audioCapturer *audio.Capturer
+	audioCapturer    *audio.Capturer
-	playbackVol   int  // 0-100
+	playbackVol      int       // 0-100
-	micLevel      int  // 0-100 (current input level)
+	micLevel         int       // 0-100 (current input level)
-	isMuted       bool // Mic muted
+	isMuted          bool      // Mic muted
-	isPTT         bool // Push-to-talk active
+	isPTT            bool      // Push-to-talk active (Manual TX)
+	vadEnabled       bool      // Voice Activation Detection active
+	vadThreshold     int       // 0-100 threshold for VAD
+	vadLastTriggered time.Time // Last time VAD threshold was exceeded
 
 	// Popup State
 	showPokePopup    bool
@@ -140,6 +143,8 @@ func NewModel(serverAddr, nickname string) *Model {
 		logMessages:    []string{"Starting..."},
 		talkingClients: make(map[uint16]bool),
 		playbackVol:    80, // Default 80% volume
+		vadEnabled:     true,
+		vadThreshold:   50,
 	}
 }
 
@@ -299,16 +304,53 @@ func (m *Model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 		} else {
 			m.audioCapturer = capturer
 			// Set callback to send audio to server when PTT is active
+			// Set callback to send audio to server when PTT is active
 			m.audioCapturer.SetCallback(func(samples []int16) {
-				if m.isPTT && m.client != nil && !m.isMuted {
+				// Calculate level of this frame for VAD decision
+				// Note: GetLevel() is smoothed, we might want instant frame level for VAD trigger?
+				// But pkg/audio/level.go is efficient. Let's re-calculate for precision.
+				level := audio.CalculateRMSLevel(samples)
+
+				// Determine if we should transmit
+				shouldTransmit := false
+
+				// Manual PTT (Locked on with 'v')
+				if m.isPTT {
+					shouldTransmit = true
+				}
+
+				// VAD Logic
+				if m.vadEnabled && !m.isMuted {
+					if level > m.vadThreshold {
+						shouldTransmit = true
+						m.vadLastTriggered = time.Now()
+					} else if !m.vadLastTriggered.IsZero() && time.Since(m.vadLastTriggered) < 1*time.Second {
+						// Hold VAD open for 1 second (decay)
+						shouldTransmit = true
+					}
+				}
+
+				// Allow transmission if forced or VAD triggered
+				if shouldTransmit && m.client != nil && !m.isMuted {
 					m.client.SendAudio(samples)
 				}
-				// Update mic level for display
+
+				// Update mic level for display (use the calculated level)
 				if m.program != nil {
-					m.program.Send(micLevelMsg(m.audioCapturer.GetLevel()))
+					// Use goroutine to prevent blocking the capture loop if the UI is busy (e.g. shutting down)
+					go m.program.Send(micLevelMsg(level))
 				}
 			})
 			m.addLog("Audio capturer initialized")
+
+			// Start capture immediately if VAD is enabled or PTT is active
+			if m.vadEnabled || m.isPTT {
+				if err := m.audioCapturer.Start(); err != nil {
+					m.addLog("Error starting audio capture: %v", err)
+				} else {
+					m.addLog("Audio capture started (VAD/PTT active)")
+				}
+			}
 		}
 
 		// Connect asynchronously
@@ -348,16 +390,7 @@ func (m *Model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 			}
 		}
 
-		// Update mic level when PTT is active (multiply for better visibility)
+		// Legacy mic level handling removed to support VAD event-driven updates
-		if m.isPTT && m.audioCapturer != nil {
-			level := m.audioCapturer.GetLevel() * 4 // Boost for visibility
-			if level > 100 {
-				level = 100
-			}
-			m.micLevel = level
-		} else {
-			m.micLevel = 0 // Reset when not transmitting
-		}
 
 		// Continue ticking (100ms for responsive mic level)
 		return m, tea.Tick(100*time.Millisecond, func(t time.Time) tea.Msg {
@@ -649,16 +682,58 @@ func (m *Model) handleKeyPress(msg tea.KeyMsg) (tea.Model, tea.Cmd) {
 			}
 			return m, nil
 
+		case "ctrl+up", "ctrl+right":
+			// Increase VAD threshold
+			m.vadThreshold += 5
+			if m.vadThreshold > 100 {
+				m.vadThreshold = 100
+			}
+			m.addLog("VAD Threshold: %d", m.vadThreshold)
+			return m, nil
+
+		case "ctrl+down", "ctrl+left":
+			// Decrease VAD threshold
+			m.vadThreshold -= 5
+			if m.vadThreshold < 0 {
+				m.vadThreshold = 0
+			}
+			m.addLog("VAD Threshold: %d", m.vadThreshold)
+			return m, nil
+
+		case "g", "G":
+			// Toggle VAD (Gate)
+			m.vadEnabled = !m.vadEnabled
+			state := "OFF"
+			if m.vadEnabled {
+				state = "ON"
+				// Ensure capturer is running if VAD is on
+				if m.audioCapturer != nil && !m.audioCapturer.IsRunning() {
+					if err := m.audioCapturer.Start(); err != nil {
+						m.addLog("Error starting VAD capture: %v", err)
+					}
+				}
+			} else {
+				// Stop if PTT is also off
+				if !m.isPTT && m.audioCapturer != nil && m.audioCapturer.IsRunning() {
+					m.audioCapturer.Stop()
+				}
+			}
+			m.addLog("Voice Activation (Gate): %s", state)
+			return m, nil
+
 		case "v", "V":
 			// Toggle voice (PTT)
 			m.isPTT = !m.isPTT
 			if m.isPTT {
-				if m.audioCapturer != nil {
+				if m.audioCapturer != nil && !m.audioCapturer.IsRunning() {
-					m.audioCapturer.Start()
+					if err := m.audioCapturer.Start(); err != nil {
+						m.addLog("Audio capture error: %v", err)
+					}
 				}
 				m.addLog("🎤 Transmitting...")
 			} else {
-				if m.audioCapturer != nil {
+				// Stop only if VAD is also off
+				if !m.vadEnabled && m.audioCapturer != nil && m.audioCapturer.IsRunning() {
 					m.audioCapturer.Stop()
 				}
 				m.addLog("🎤 Stopped transmitting")
@@ -1010,7 +1085,7 @@ func (m *Model) View() string {
 	if m.showLog {
 		logHelp = "L chat"
 	}
-	help := lipgloss.NewStyle().Faint(true).Render(fmt.Sprintf("↑↓ navigate │ Enter join │ Tab switch │ %s │ V talk │ M mute │ +/- vol │ q quit", logHelp))
+	help := lipgloss.NewStyle().Faint(true).Render(fmt.Sprintf("↑↓ nav │ Ent join │ Tab switch │ %s │ V PTT │ G VAD │ ^↕↔ thresh │ M mute │ +/- vol │ q quit", logHelp))
 
 	// Combine panels
 	panels := lipgloss.JoinHorizontal(lipgloss.Top, channelPanel, rightPanel)
@@ -1053,12 +1128,85 @@ func (m *Model) renderStatusBar() string {
 	}
 	volPart := fmt.Sprintf("%s:%s%d%%", muteIcon, volBar, m.playbackVol)
 
-	micBar := audio.LevelToBar(m.micLevel, 6)
+	// Custom Mic Bar with VAD Threshold
-	pttIcon := "MIC"
+	micBarWidth := 8
-	if m.isPTT {
+	var micBar string
-		pttIcon = "*TX*"
+
+	if m.vadEnabled {
+		// Calculate threshold position
+		threshPos := m.vadThreshold * micBarWidth / 100
+		if threshPos >= micBarWidth {
+			threshPos = micBarWidth - 1
+		}
+
+		// Calculate filled position based on current level
+		filled := m.micLevel * micBarWidth / 100
+
+		// Build bar
+		var sb strings.Builder
+		for i := 0; i < micBarWidth; i++ {
+			char := "░"
+			if i < filled {
+				char = "█"
+			}
+
+			// Overlay threshold marker
+			if i == threshPos {
+				if i < filled {
+					// Threshold is met
+					char = "▓"
+				} else {
+					// Threshold not met
+					char = "│"
+				}
+			}
+			sb.WriteString(char)
+		}
+		micBar = sb.String()
+	} else {
+		micBar = audio.LevelToBar(m.micLevel, micBarWidth)
 	}
-	micPart := fmt.Sprintf("%s:%s", pttIcon, micBar)
+
+	pttStyle := lipgloss.NewStyle()
+	pttIcon := "MIC"
+
+	if m.isPTT {
+		pttIcon = " ON"                                                  // Manual ON
+		pttStyle = pttStyle.Foreground(lipgloss.Color("196")).Bold(true) // Red
+	} else if m.vadEnabled {
+		pttIcon = "VAD"
+		// Check if actively transmitting (using logic with decay)
+		isTransmitting := false
+		if !m.isMuted {
+			if m.micLevel > m.vadThreshold {
+				isTransmitting = true
+			} else if !m.vadLastTriggered.IsZero() && time.Since(m.vadLastTriggered) < 1*time.Second {
+				isTransmitting = true
+			}
+		}
+
+		if isTransmitting {
+			// Transmitting via VAD: Red/Bold
+			pttStyle = pttStyle.Foreground(lipgloss.Color("196")).Bold(true)
+		} else {
+			// Idle VAD: Gray/Faint
+			pttStyle = pttStyle.Foreground(lipgloss.Color("240")).Faint(true)
+		}
+	} else {
+		// Standard Mic (PTT mode but off)
+		pttStyle = pttStyle.Foreground(lipgloss.Color("255")) // White
+	}
+
+	// Apply status bar background color to prevent cutting
+	pttStyle = pttStyle.Background(lipgloss.Color("57")) // Matches Top Bar Background
+
+	// Style for the bar itself to maintain background continuity
+	barStyle := lipgloss.NewStyle().Background(lipgloss.Color("57")).Foreground(lipgloss.Color("255"))
+
+	micPart := fmt.Sprintf("%s%s%s",
+		pttStyle.Render(pttIcon),
+		barStyle.Render(":"),
+		barStyle.Render(micBar))
 
 	rightPart := fmt.Sprintf("%s | %s ", volPart, micPart)
 

pkg/audio/capture.go

@@ -22,6 +22,7 @@ type Capturer struct {
 	running       bool
 	mu            sync.Mutex
 	stopChan      chan struct{}
+	wg            sync.WaitGroup
 
 	// Callback for captured audio (called with 960-sample frames)
 	onAudio func(samples []int16)
@@ -136,6 +137,7 @@ func (c *Capturer) Start() error {
 		return fmt.Errorf("failed to start audio client: %w", err)
 	}
 
+	c.wg.Add(1)
 	go c.captureLoop()
 	return nil
 }
@@ -151,6 +153,7 @@ func (c *Capturer) Stop() {
 	c.mu.Unlock()
 
 	close(c.stopChan)
+	c.wg.Wait() // Wait for capture loop to finish before proceeding
 	c.client.Stop()
 }
 
@@ -180,6 +183,7 @@ func (c *Capturer) IsRunning() bool {
 }
 
 func (c *Capturer) captureLoop() {
+	defer c.wg.Done()
 	ticker := time.NewTicker(10 * time.Millisecond) // Check more often than 20ms
 	defer ticker.Stop()
 

pkg/audio/level.go

@@ -4,7 +4,7 @@ import (
 	"math"
 )
 
-// CalculateRMSLevel calculates the RMS level of PCM samples and returns 0-100
+// CalculateRMSLevel calculates the RMS level of PCM samples and returns 0-100 (Logarithmic/dB)
 func CalculateRMSLevel(samples []int16) int {
 	if len(samples) == 0 {
 		return 0
@@ -16,15 +16,33 @@ func CalculateRMSLevel(samples []int16) int {
 	}
 
 	rms := math.Sqrt(sum / float64(len(samples)))
-	// Normalize to 0-100 (max int16 is 32767)
+
-	level := int(rms / 32767.0 * 100.0)
+	// Normalize to 0.0 - 1.0
+	val := rms / 32768.0
+	if val < 0.000001 { // Avoid log(0)
+		return 0
+	}
+
+	// Convert to dB
+	db := 20 * math.Log10(val)
+
+	// Map -50dB (silence floor) to 0dB (max) to 0-100
+	const minDB = -50.0
+
+	if db < minDB {
+		return 0
+	}
+
+	// Scale
+	level := int((db - minDB) * (100.0 / (0 - minDB)))
+
 	if level > 100 {
 		level = 100
 	}
 	return level
 }
 
-// CalculatePeakLevel returns the peak level of PCM samples as 0-100
+// CalculatePeakLevel returns the peak level of PCM samples as 0-100 (Logarithmic/dB)
 func CalculatePeakLevel(samples []int16) int {
 	if len(samples) == 0 {
 		return 0
@@ -40,7 +58,26 @@ func CalculatePeakLevel(samples []int16) int {
 		}
 	}
 
-	return int(float64(peak) / 32767.0 * 100.0)
+	// Normalize
+	val := float64(peak) / 32768.0
+	if val < 0.000001 {
+		return 0
+	}
+
+	db := 20 * math.Log10(val)
+	const minDB = -50.0
+
+	if db < minDB {
+		// Linear falloff for very low signals to avoid clutter
+		return 0
+	}
+
+	level := int((db - minDB) * (100.0 / (0 - minDB)))
+	if level > 100 {
+		level = 100
+	}
+
+	return level
 }
 
 // LevelToBar converts a 0-100 level to a visual bar string