go-ts/cmd/voicebot/main.go

package main

import (
	"flag"
	"log"
	"os"
	"os/signal"
	"strings"
	"sync"
	"syscall"
	"time"

	"go-ts/pkg/ts3client"
	"go-ts/pkg/xai"
)

// VoiceSession represents an active xAI voice session for a user
type VoiceSession struct {
	ClientID    uint16
	Nickname    string
	XAI         *xai.Client
	AudioBuffer []int16       // Buffer to accumulate audio samples
	AudioQueue  chan []int16  // Queue for sending audio with proper timing
	done        chan struct{} // Signal to stop audio sender
}

// Bot manages the TeamSpeak connection and xAI sessions
type Bot struct {
	ts3      *ts3client.Client
	apiKey   string
	voice    string
	prompt   string
	greeting string // Optional greeting when user joins

	selfID     uint16 // Our own ClientID
	sessions   map[uint16]*VoiceSession
	sessionsMu sync.RWMutex
	startTime  time.Time
}

func main() {
	serverAddr := flag.String("server", "127.0.0.1:9987", "TeamSpeak 3 Server Address")
	nickname := flag.String("nickname", "GrokBot", "Bot nickname")
	voice := flag.String("voice", xai.VoiceAra, "xAI voice (Ara, Rex, Sal, Eve, Leo)")
	greeting := flag.String("greeting", "Saluda brevemente al usuario que acaba de unirse.", "Greeting message (empty to disable)")
	flag.Parse()

	apiKey := os.Getenv("XAI_API_KEY")
	if apiKey == "" {
		log.Fatal("XAI_API_KEY environment variable not set")
	}

	log.Println("=== xAI Voice Bot for TeamSpeak ===")
	log.Printf("Server: %s", *serverAddr)
	log.Printf("Nickname: %s", *nickname)
	log.Printf("Voice: %s", *voice)

	bot := &Bot{
		apiKey:    apiKey,
		voice:     *voice,
		prompt:    "Eres Grok, un asistente de voz amigable y útil. Responde de forma concisa y natural.",
		greeting:  *greeting,
		sessions:  make(map[uint16]*VoiceSession),
		startTime: time.Now(),
	}

	// Create TeamSpeak client
	bot.ts3 = ts3client.New(*serverAddr, ts3client.Config{
		Nickname: *nickname,
	})

	// Register event handlers
	bot.ts3.On(ts3client.EventConnected, func(e *ts3client.ConnectedEvent) {
		bot.selfID = e.ClientID // Store our own ID
		log.Printf("✓ Conectado a TeamSpeak! ClientID=%d, Server=%s", e.ClientID, e.ServerName)
	})

	bot.ts3.On(ts3client.EventChannelList, func(e *ts3client.ChannelListEvent) {
		log.Printf("✓ %d canales disponibles", len(e.Channels))
	})

	bot.ts3.On(ts3client.EventClientEnter, func(e *ts3client.ClientEnterEvent) {
		log.Printf("→ Usuario entró: %s (ID=%d)", e.Nickname, e.ClientID)

		// Don't create session for ourselves (compare by ID, not nickname)
		if e.ClientID == bot.selfID {
			log.Printf("   (Soy yo, ignorando)")
			return
		}

		// Create xAI session for this user
		go bot.createSession(e.ClientID, e.Nickname, bot.greeting)
	})

	bot.ts3.On(ts3client.EventClientLeft, func(e *ts3client.ClientLeftEvent) {
		log.Printf("← Usuario salió: ID=%d (%s)", e.ClientID, e.Reason)

		// Close xAI session for this user
		bot.closeSession(e.ClientID)
	})

	bot.ts3.On(ts3client.EventAudio, func(e *ts3client.AudioEvent) {
		// Forward audio from TeamSpeak to all xAI sessions
		// Forward audio ONLY to the sender's session
		bot.sessionsMu.RLock()
		if session, ok := bot.sessions[e.SenderID]; ok {
			if session.XAI != nil && session.XAI.IsConnected() {
				session.XAI.SendAudio(e.PCM)
			}
		}
		bot.sessionsMu.RUnlock()
	})

	bot.ts3.On(ts3client.EventError, func(e *ts3client.ErrorEvent) {
		if e.ID != "0" {
			log.Printf("! Error del servidor: [%s] %s", e.ID, e.Message)
		}
	})

	// Handle shutdown
	shutdownDone := make(chan struct{})
	go func() {
		sigChan := make(chan os.Signal, 1)
		signal.Notify(sigChan, syscall.SIGINT, syscall.SIGTERM)
		<-sigChan
		log.Println("Cerrando...")

		// Close all xAI sessions and audio senders first
		bot.sessionsMu.Lock()
		for _, session := range bot.sessions {
			// Close audio sender first
			select {
			case <-session.done:
				// Already closed
			default:
				close(session.done)
			}
			// Then close xAI
			if session.XAI != nil {
				session.XAI.Close()
			}
		}
		bot.sessionsMu.Unlock()

		// Wait for audio senders to stop
		time.Sleep(200 * time.Millisecond)

		// Now disconnect from TeamSpeak
		bot.ts3.Disconnect()

		// Signal main that we are done
		close(shutdownDone)
	}()

	// Start global audio mixer
	go bot.runAudioMixer(shutdownDone)

	// Connect to TeamSpeak
	if err := bot.ts3.Connect(); err != nil {
		// If connect returns error, check if it's because we're shutting down
		select {
		case <-shutdownDone:
			// Normal shutdown
			log.Println("Conexión cerrada por shutdown")
		default:
			log.Fatalf("Error de conexión: %v", err)
		}
	}

	// Wait for shutdown to complete if we returned from Connect cleanly
	log.Println("Esperando confirmación final de shutdown...")
	<-shutdownDone
	log.Println("Shutdown completado. Saliendo.")
	os.Exit(0)
}

// createSession creates a new xAI voice session for a user
func (b *Bot) createSession(clientID uint16, nickname string, greeting string) {
	log.Printf("[Session] Creando sesión xAI para %s...", nickname)

	// Create session with audio queue
	session := &VoiceSession{
		ClientID:    clientID,
		Nickname:    nickname,
		AudioBuffer: make([]int16, 0, 960*10),
		AudioQueue:  make(chan []int16, 500), // Buffer up to 500 frames (~10 sec)
		done:        make(chan struct{}),
	}

	// Start audio sender goroutine with proper 20ms timing
	// Global audio mixer handles sending
	// go b.audioSender(session)

	xaiClient := xai.New(b.apiKey)

	// Set up audio callback - buffer and queue in 960-sample chunks
	xaiClient.OnAudio(func(pcm []int16) {
		b.sessionsMu.Lock()
		session.AudioBuffer = append(session.AudioBuffer, pcm...)

		// Queue complete 960-sample frames
		for len(session.AudioBuffer) >= 960 {
			frame := make([]int16, 960)
			copy(frame, session.AudioBuffer[:960])
			session.AudioBuffer = session.AudioBuffer[960:]

			// Non-blocking send to queue
			select {
			case session.AudioQueue <- frame:
			default:
				// Queue full, drop frame
			}
		}
		b.sessionsMu.Unlock()
	})

	// Set up transcript callback for logging
	xaiClient.OnTranscript(func(text string) {
		log.Printf("[Grok] %s", text)
	})

	// Clear audio queue when user starts speaking (interruption)
	xaiClient.OnSpeechStarted(func() {
		// Disable queue clearing for now to prevent cutting off greetings due to sensitive VAD
		/*
			b.sessionsMu.Lock()
			// Clear the buffer
			session.AudioBuffer = session.AudioBuffer[:0]
			// Drain the queue
			for len(session.AudioQueue) > 0 {
				<-session.AudioQueue
			}
			b.sessionsMu.Unlock()
		*/
		log.Printf("[Session] Speech started by %s (VAD) - Ignoring interruption to ensure playback", nickname)
	})

	// Connect to xAI
	if err := xaiClient.Connect(); err != nil {
		log.Printf("[Session] Error conectando a xAI: %v", err)
		close(session.done)
		return
	}

	// Configure the session
	if err := xaiClient.ConfigureSession(b.voice, b.prompt); err != nil {
		log.Printf("[Session] Error configurando sesión: %v", err)
		xaiClient.Close()
		close(session.done)
		return
	}

	// Store the xAI client in session
	session.XAI = xaiClient

	b.sessionsMu.Lock()
	b.sessions[clientID] = session
	b.sessionsMu.Unlock()

	log.Printf("[Session] ✓ Sesión xAI activa para %s", nickname)

	// Send greeting to start conversation (if configured)
	// Send greeting to start conversation (if configured)
	if strings.TrimSpace(greeting) != "" {
		// Only greet if we are past the startup grace period (3 seconds)
		// This prevents "Greeting Storm" when joining a channel with existing users
		if time.Since(b.startTime) > 3*time.Second {
			go func() {
				time.Sleep(500 * time.Millisecond) // Small delay for session to stabilize
				if err := xaiClient.SendText(greeting); err != nil {
					log.Printf("[Session] Error enviando saludo: %v", err)
				}
			}()
		} else {
			log.Printf("[Session] Omitiendo saludo inicial para %s (sesión existente detectada en arranque)", nickname)
		}
	}
}

// runAudioMixer mixes audio from all active sessions and sends it to TeamSpeak
func (b *Bot) runAudioMixer(stop <-chan struct{}) {
	ticker := time.NewTicker(20 * time.Millisecond)
	defer ticker.Stop()

	mixedFrame := make([]int16, 960)

	for {
		select {
		case <-stop:
			return
		case <-ticker.C:
			hasAudio := false

			// Zero output buffer
			for i := range mixedFrame {
				mixedFrame[i] = 0
			}

			b.sessionsMu.RLock()
			for _, session := range b.sessions {
				// Try to get a frame from the queue
				select {
				case frame := <-session.AudioQueue:
					hasAudio = true
					// Mix (Sum and Clamp)
					for i := 0; i < 960; i++ {
						if i >= len(frame) {
							break
						}
						val := int32(mixedFrame[i]) + int32(frame[i])
						if val > 32767 {
							val = 32767
						}
						if val < -32768 {
							val = -32768
						}
						mixedFrame[i] = int16(val)
					}
				default:
					// No audio from this session
				}
			}
			b.sessionsMu.RUnlock()

			if hasAudio {
				if err := b.ts3.SendAudio(mixedFrame); err != nil {
					log.Printf("[Mixer] Error sending audio: %v", err)
				}
			}
		}
	}
}

// closeSession closes an xAI session for a user
func (b *Bot) closeSession(clientID uint16) {
	b.sessionsMu.Lock()
	defer b.sessionsMu.Unlock()

	if session, ok := b.sessions[clientID]; ok {
		log.Printf("[Session] Cerrando sesión xAI para %s", session.Nickname)
		if session.XAI != nil {
			session.XAI.Close()
		}
		delete(b.sessions, clientID)
	}
}