init commit

go.mod

@@ -22,6 +22,8 @@ require (
 	github.com/mymmrac/telego v1.7.0
 	github.com/open-dingtalk/dingtalk-stream-sdk-go v0.9.1
 	github.com/openai/openai-go/v3 v3.22.0
+	github.com/pion/rtp v1.8.7
+	github.com/pion/webrtc/v3 v3.3.6
 	github.com/rivo/tview v0.42.0
 	github.com/rs/zerolog v1.34.0
 	github.com/slack-go/slack v0.17.3
@@ -53,6 +55,7 @@ require (
 	github.com/mattn/go-isatty v0.0.20 // indirect
 	github.com/ncruces/go-strftime v1.0.0 // indirect
 	github.com/petermattis/goid v0.0.0-20260226131333-17d1149c6ac6 // indirect
+	github.com/pion/randutil v0.1.0 // indirect
 	github.com/pmezard/go-difflib v1.0.0 // indirect
 	github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec // indirect
 	github.com/rivo/uniseg v0.4.7 // indirect

go.sum

@@ -162,6 +162,12 @@ github.com/openai/openai-go/v3 v3.22.0 h1:6MEoNoV8sbjOVmXdvhmuX3BjVbVdcExbVyGixi
 github.com/openai/openai-go/v3 v3.22.0/go.mod h1:cdufnVK14cWcT9qA1rRtrXx4FTRsgbDPW7Ia7SS5cZo=
 github.com/petermattis/goid v0.0.0-20260226131333-17d1149c6ac6 h1:rh2lKw/P/EqHa724vYH2+VVQ1YnW4u6EOXl0PMAovZE=
 github.com/petermattis/goid v0.0.0-20260226131333-17d1149c6ac6/go.mod h1:pxMtw7cyUw6B2bRH0ZBANSPg+AoSud1I1iyJHI69jH4=
+github.com/pion/randutil v0.1.0 h1:CFG1UdESneORglEsnimhUjf33Rwjubwj6xfiOXBa3mA=
+github.com/pion/randutil v0.1.0/go.mod h1:XcJrSMMbbMRhASFVOlj/5hQial/Y8oH/HVo7TBZq+j8=
+github.com/pion/rtp v1.8.7 h1:qslKkG8qxvQ7hqaxkmL7Pl0XcUm+/Er7nMnu6Vq+ZxM=
+github.com/pion/rtp v1.8.7/go.mod h1:pBGHaFt/yW7bf1jjWAoUjpSNoDnw98KTMg+jWWvziqU=
+github.com/pion/webrtc/v3 v3.3.6 h1:7XAh4RPtlY1Vul6/GmZrv7z+NnxKA6If0KStXBI2ZLE=
+github.com/pion/webrtc/v3 v3.3.6/go.mod h1:zyN7th4mZpV27eXybfR/cnUf3J2DRy8zw/mdjD9JTNM=
 github.com/pkg/diff v0.0.0-20210226163009-20ebb0f2a09e/go.mod h1:pJLUxLENpZxwdsKMEsNbx1VGcRFpLqf3715MtcvvzbA=
 github.com/pkg/errors v0.9.1/go.mod h1:bwawxfHBFNV+L2hUp1rHADufV3IMtnDRdf1r5NINEl0=
 github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=

pkg/bus/bus.go

@@ -34,6 +34,8 @@ type MessageBus struct {
 	inbound       chan InboundMessage
 	outbound      chan OutboundMessage
 	outboundMedia chan OutboundMediaMessage
+	audioChunks   chan AudioChunk
+	voiceControls chan VoiceControl
 
 	closeOnce      sync.Once
 	done           chan struct{}
@@ -47,6 +49,8 @@ func NewMessageBus() *MessageBus {
 		inbound:       make(chan InboundMessage, defaultBusBufferSize),
 		outbound:      make(chan OutboundMessage, defaultBusBufferSize),
 		outboundMedia: make(chan OutboundMediaMessage, defaultBusBufferSize),
+		audioChunks:   make(chan AudioChunk, defaultBusBufferSize*4), // Audio chunks need more buffer
+		voiceControls: make(chan VoiceControl, defaultBusBufferSize),
 		done:          make(chan struct{}),
 	}
 }
@@ -103,6 +107,22 @@ func (mb *MessageBus) OutboundMediaChan() <-chan OutboundMediaMessage {
 	return mb.outboundMedia
 }
 
+func (mb *MessageBus) PublishAudioChunk(ctx context.Context, chunk AudioChunk) error {
+	return publish(ctx, mb, mb.audioChunks, chunk)
+}
+
+func (mb *MessageBus) AudioChunksChan() <-chan AudioChunk {
+	return mb.audioChunks
+}
+
+func (mb *MessageBus) PublishVoiceControl(ctx context.Context, ctrl VoiceControl) error {
+	return publish(ctx, mb, mb.voiceControls, ctrl)
+}
+
+func (mb *MessageBus) VoiceControlsChan() <-chan VoiceControl {
+	return mb.voiceControls
+}
+
 // SetStreamDelegate registers a StreamDelegate (typically the channel Manager).
 func (mb *MessageBus) SetStreamDelegate(d StreamDelegate) {
 	mb.streamDelegate.Store(d)
@@ -132,6 +152,8 @@ func (mb *MessageBus) Close() {
 		close(mb.inbound)
 		close(mb.outbound)
 		close(mb.outboundMedia)
+		close(mb.audioChunks)
+		close(mb.voiceControls)
 
 		// clean up any remaining messages in channels
 		drained := 0
@@ -144,6 +166,12 @@ func (mb *MessageBus) Close() {
 		for range mb.outboundMedia {
 			drained++
 		}
+		for range mb.audioChunks {
+			drained++
+		}
+		for range mb.voiceControls {
+			drained++
+		}
 
 		if drained > 0 {
 			logger.DebugCF("bus", "Drained buffered messages during close", map[string]any{

pkg/bus/types.go

@@ -51,3 +51,23 @@ type OutboundMediaMessage struct {
 	ChatID  string      `json:"chat_id"`
 	Parts   []MediaPart `json:"parts"`
 }
+
+// AudioChunk represents a chunk of streaming voice data.
+type AudioChunk struct {
+	SessionID  string `json:"session_id"`
+	SpeakerID  string `json:"speaker_id"` // User ID or SSRC
+	ChatID     string `json:"chat_id"`    // Where to respond
+	Sequence   uint64 `json:"sequence"`
+	Timestamp  uint32 `json:"timestamp"`
+	SampleRate int    `json:"sample_rate"`
+	Channels   int    `json:"channels"`
+	Format     string `json:"format"` // "opus", "pcm", etc
+	Data       []byte `json:"data"`
+}
+
+// VoiceControl represents state or commands for voice sessions.
+type VoiceControl struct {
+	SessionID string `json:"session_id"`
+	Type      string `json:"type"`  // "state", "command"
+	Action    string `json:"action"` // "idle", "listening", "start", "stop"
+}

pkg/channels/discord/discord.go

@@ -42,6 +42,7 @@ type DiscordChannel struct {
 	typingMu   sync.Mutex
 	typingStop map[string]chan struct{} // chatID → stop signal
 	botUserID  string                   // stored for mention checking
+	bus        *bus.MessageBus
 }
 
 func NewDiscordChannel(cfg config.DiscordConfig, bus *bus.MessageBus) (*DiscordChannel, error) {
@@ -73,6 +74,7 @@ func NewDiscordChannel(cfg config.DiscordConfig, bus *bus.MessageBus) (*DiscordC
 		config:      cfg,
 		ctx:         context.Background(),
 		typingStop:  make(map[string]chan struct{}),
+		bus:         bus,
 	}, nil
 }
 
@@ -321,6 +323,10 @@ func (c *DiscordChannel) handleMessage(s *discordgo.Session, m *discordgo.Messag
 		return
 	}
 
+	if c.handleVoiceCommand(s, m) {
+		return
+	}
+
 	// Check allowlist first to avoid downloading attachments for rejected users
 	sender := bus.SenderInfo{
 		Platform:    "discord",

pkg/channels/discord/voice.go

@@ -0,0 +1,86 @@
+package discord
+
+import (
+	"fmt"
+
+	"github.com/bwmarrin/discordgo"
+	"github.com/sipeed/picoclaw/pkg/bus"
+	"github.com/sipeed/picoclaw/pkg/logger"
+)
+
+func (c *DiscordChannel) handleVoiceCommand(s *discordgo.Session, m *discordgo.MessageCreate) bool {
+	if m.Content == "!vc join" {
+		vs, err := s.State.VoiceState(m.GuildID, m.Author.ID)
+		if err != nil || vs == nil {
+			s.ChannelMessageSend(m.ChannelID, "You need to be in a voice channel first!")
+			return true
+		}
+
+		logger.InfoCF("discord", "Joining voice channel", map[string]any{"channel": vs.ChannelID})
+		vc, err := s.ChannelVoiceJoin(m.GuildID, vs.ChannelID, false, false)
+		if err != nil {
+			s.ChannelMessageSend(m.ChannelID, fmt.Sprintf("Failed to join voice channel: %v", err))
+			return true
+		}
+
+		go c.receiveVoice(vc, m.GuildID, m.ChannelID)
+		s.ChannelMessageSend(m.ChannelID, "Joined Voice Channel! Listening for audio...")
+		return true
+	} else if m.Content == "!vc leave" {
+		vc, exists := s.VoiceConnections[m.GuildID]
+		if exists && vc != nil {
+			vc.Disconnect()
+			s.ChannelMessageSend(m.ChannelID, "Left Voice Channel.")
+		} else {
+			s.ChannelMessageSend(m.ChannelID, "Not in a voice channel.")
+		}
+		return true
+	}
+	return false
+}
+
+func (c *DiscordChannel) receiveVoice(vc *discordgo.VoiceConnection, guildID string, chatID string) {
+	logger.InfoCF("discord", "Started listening for voice", map[string]any{"guild": guildID})
+
+	sessionID := fmt.Sprintf("discord_vc_%s", guildID)
+
+	c.bus.PublishVoiceControl(c.ctx, bus.VoiceControl{
+		SessionID: sessionID,
+		Type:      "state",
+		Action:    "listening",
+	})
+
+	var sequence uint64 = 0
+
+	for {
+		select {
+		case <-c.ctx.Done():
+			return
+		case p, ok := <-vc.OpusRecv:
+			if !ok {
+				logger.InfoCF("discord", "Voice channel closed", map[string]any{"guild": guildID})
+				return
+			}
+
+			if p == nil {
+				continue
+			}
+
+			sequence++
+
+			chunk := bus.AudioChunk{
+				SessionID:  sessionID,
+				SpeakerID:  fmt.Sprintf("%d", p.SSRC),
+				ChatID:     chatID,
+				Sequence:   sequence,
+				Timestamp:  p.Timestamp,
+				SampleRate: 48000,
+				Channels:   2,
+				Format:     "opus",
+				Data:       p.Opus,
+			}
+
+			c.bus.PublishAudioChunk(c.ctx, chunk)
+		}
+	}
+}

pkg/gateway/gateway.go

@@ -60,6 +60,7 @@ type services struct {
 	ChannelManager   *channels.Manager
 	DeviceService    *devices.Service
 	HealthServer     *health.Server
+	VoiceAgentCancel context.CancelFunc
 	manualReloadChan chan struct{}
 	reloading        atomic.Bool
 }
@@ -305,6 +306,12 @@ func setupAndStartServices(
 	if transcriber := voice.DetectTranscriber(cfg); transcriber != nil {
 		agentLoop.SetTranscriber(transcriber)
 		logger.InfoCF("voice", "Transcription enabled (agent-level)", map[string]any{"provider": transcriber.Name()})
+
+		// Start Voice Agent Orchestrator
+		vaCtx, vaCancel := context.WithCancel(context.Background())
+		runningServices.VoiceAgentCancel = vaCancel
+		voiceAgent := voice.NewAgent(msgBus, transcriber)
+		voiceAgent.Start(vaCtx)
 	}
 
 	enabledChannels := runningServices.ChannelManager.GetEnabledChannels()
@@ -351,6 +358,9 @@ func stopAndCleanupServices(runningServices *services, shutdownTimeout time.Dura
 	if !isReload && runningServices.ChannelManager != nil {
 		runningServices.ChannelManager.StopAll(shutdownCtx)
 	}
+	if runningServices.VoiceAgentCancel != nil {
+		runningServices.VoiceAgentCancel()
+	}
 	if runningServices.DeviceService != nil {
 		runningServices.DeviceService.Stop()
 	}
@@ -532,6 +542,12 @@ func restartServices(
 	al.SetTranscriber(transcriber)
 	if transcriber != nil {
 		logger.InfoCF("voice", "Transcription re-enabled (agent-level)", map[string]any{"provider": transcriber.Name()})
+
+		// Start Voice Agent Orchestrator on reload
+		vaCtx, vaCancel := context.WithCancel(context.Background())
+		runningServices.VoiceAgentCancel = vaCancel
+		voiceAgent := voice.NewAgent(msgBus, transcriber)
+		voiceAgent.Start(vaCtx)
 	} else {
 		logger.InfoCF("voice", "Transcription disabled", nil)
 	}

pkg/voice/agent.go

@@ -0,0 +1,195 @@
+package voice
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"path/filepath"
+	"sync"
+	"time"
+
+	"github.com/pion/rtp"
+	"github.com/pion/webrtc/v3/pkg/media/oggwriter"
+	"github.com/sipeed/picoclaw/pkg/bus"
+	"github.com/sipeed/picoclaw/pkg/logger"
+)
+
+type speechAccumulator struct {
+	writer      *oggwriter.OggWriter
+	file        string
+	lastAudioAt time.Time
+	mu          sync.Mutex
+	closed      bool
+	chatID      string
+	speakerID   string
+}
+
+func (a *speechAccumulator) Push(chunk bus.AudioChunk) {
+	a.mu.Lock()
+	defer a.mu.Unlock()
+
+	if a.closed {
+		return
+	}
+
+	a.lastAudioAt = time.Now()
+
+	pkt := &rtp.Packet{
+		Header: rtp.Header{
+			SequenceNumber: uint16(chunk.Sequence),
+			Timestamp:      chunk.Timestamp,
+			SSRC:           uint32(chunk.Sequence), // Arbitrary dummy
+		},
+		Payload: chunk.Data,
+	}
+
+	if err := a.writer.WriteRTP(pkt); err != nil {
+		logger.ErrorCF("voice-agent", "Failed to write RTP", map[string]any{"error": err})
+	}
+}
+
+func (a *speechAccumulator) Close() {
+	a.mu.Lock()
+	defer a.mu.Unlock()
+	if !a.closed {
+		a.writer.Close()
+		a.closed = true
+	}
+}
+
+type Agent struct {
+	bus         *bus.MessageBus
+	transcriber Transcriber
+
+	mu       sync.Mutex
+	sessions map[string]*speechAccumulator // keyed by sessionID_speakerID
+}
+
+func NewAgent(mb *bus.MessageBus, t Transcriber) *Agent {
+	return &Agent{
+		bus:         mb,
+		transcriber: t,
+		sessions:    make(map[string]*speechAccumulator),
+	}
+}
+
+func (a *Agent) Start(ctx context.Context) {
+	logger.InfoCF("voice-agent", "Started Voice Agent orchestrator", nil)
+	go a.listenChunks(ctx)
+	go a.vadTick(ctx)
+}
+
+func (a *Agent) listenChunks(ctx context.Context) {
+	chunks := a.bus.AudioChunksChan()
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case chunk := <-chunks:
+			a.handleChunk(chunk)
+		}
+	}
+}
+
+func (a *Agent) handleChunk(chunk bus.AudioChunk) {
+	a.mu.Lock()
+	defer a.mu.Unlock()
+
+	key := fmt.Sprintf("%s_%s", chunk.SessionID, chunk.SpeakerID)
+
+	acc, exists := a.sessions[key]
+	if !exists {
+		filename := filepath.Join(os.TempDir(), fmt.Sprintf("voice_%s_%d.ogg", key, time.Now().UnixNano()))
+		writer, err := oggwriter.New(filename, uint32(chunk.SampleRate), uint16(chunk.Channels))
+		if err != nil {
+			logger.ErrorCF("voice-agent", "Failed to create OggWriter", map[string]any{"error": err})
+			return
+		}
+
+		acc = &speechAccumulator{
+			writer:      writer,
+			file:        filename,
+			lastAudioAt: time.Now(),
+			chatID:      chunk.ChatID,
+			speakerID:   chunk.SpeakerID,
+		}
+		a.sessions[key] = acc
+		logger.DebugCF("voice-agent", "Started accumulating voice", map[string]any{"key": key, "file": filename})
+	}
+
+	acc.Push(chunk)
+}
+
+func (a *Agent) vadTick(ctx context.Context) {
+	ticker := time.NewTicker(500 * time.Millisecond)
+	defer ticker.Stop()
+
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case <-ticker.C:
+			a.checkSilence(ctx)
+		}
+	}
+}
+
+func (a *Agent) checkSilence(ctx context.Context) {
+	a.mu.Lock()
+	now := time.Now()
+	var finished []*speechAccumulator
+
+	for key, acc := range a.sessions {
+		acc.mu.Lock()
+		last := acc.lastAudioAt
+		acc.mu.Unlock()
+
+		if now.Sub(last) > 1500*time.Millisecond {
+			acc.Close()
+			delete(a.sessions, key)
+			finished = append(finished, acc)
+		}
+	}
+	a.mu.Unlock()
+
+	for _, acc := range finished {
+		go a.processUtterance(ctx, acc)
+	}
+}
+
+func (a *Agent) processUtterance(ctx context.Context, acc *speechAccumulator) {
+	defer os.Remove(acc.file)
+
+	logger.InfoCF("voice-agent", "User finished speaking, transcribing...", map[string]any{"file": acc.file})
+
+	if a.transcriber == nil {
+		logger.ErrorCF("voice-agent", "No STT configured!", nil)
+		return
+	}
+
+	res, err := a.transcriber.Transcribe(ctx, acc.file)
+	if err != nil {
+		logger.ErrorCF("voice-agent", "Transcription failed", map[string]any{"error": err})
+		return
+	}
+
+	if res.Text == "" {
+		logger.DebugCF("voice-agent", "Ignored empty transcription", map[string]any{"file": acc.file})
+		return
+	}
+
+	logger.InfoCF("voice-agent", "Transcription result", map[string]any{"text": res.Text, "duration": res.Duration})
+
+	channelType := "discord"
+
+	a.bus.PublishInbound(ctx, bus.InboundMessage{
+		Channel:  channelType,
+		SenderID: acc.speakerID,
+		ChatID:   acc.chatID,
+		Content:  res.Text,
+		Peer:     bus.Peer{Kind: "channel", ID: acc.chatID},
+		Metadata: map[string]string{
+			"is_voice": "true",
+		},
+	})
+}

pkg/voice/transcriber.go

@@ -1,11 +1,21 @@
 package voice
 
 import (
+	"bytes"
 	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"mime/multipart"
+	"net/http"
+	"os"
+	"path/filepath"
 	"strings"
+	"time"
 
 	"github.com/sipeed/picoclaw/pkg/config"
-	"github.com/sipeed/picoclaw/pkg/providers"
+	"github.com/sipeed/picoclaw/pkg/logger"
+	"github.com/sipeed/picoclaw/pkg/utils"
 )
 
 type Transcriber interface {
@@ -13,34 +23,157 @@ type Transcriber interface {
 	Transcribe(ctx context.Context, audioFilePath string) (*TranscriptionResponse, error)
 }
 
+type GroqTranscriber struct {
+	apiKey     string
+	apiBase    string
+	httpClient *http.Client
+}
+
 type TranscriptionResponse struct {
 	Text     string  `json:"text"`
 	Language string  `json:"language,omitempty"`
 	Duration float64 `json:"duration,omitempty"`
 }
 
-func supportsAudioTranscription(model string) bool {
-	protocol, _ := providers.ExtractProtocol(model)
+func NewGroqTranscriber(apiKey string) *GroqTranscriber {
+	logger.DebugCF("voice", "Creating Groq transcriber", map[string]any{"has_api_key": apiKey != ""})
 
-	switch protocol {
-	case "openai", "azure", "azure-openai",
-		"litellm", "openrouter", "groq", "zhipu", "gemini", "nvidia",
-		"ollama", "moonshot", "shengsuanyun", "deepseek", "cerebras",
-		"vivgrid", "volcengine", "vllm", "qwen", "qwen-intl", "qwen-international", "dashscope-intl",
-		"qwen-us", "dashscope-us", "mistral", "avian", "minimax", "longcat", "modelscope", "novita",
-		"coding-plan", "alibaba-coding", "qwen-coding":
-		// These protocols all go through the OpenAI-compatible or Azure provider path in
-		// providers.CreateProviderFromConfig, so they are the only ones that can supply
-		// the audio media payload shape expected by NewAudioModelTranscriber.
-
-		// TODO: Further restrict this by modelID, since not every model under these
-		// protocols supports audio transcription.
-		return true
-	default:
-		return false
+	apiBase := "https://api.groq.com/openai/v1"
+	return &GroqTranscriber{
+		apiKey:  apiKey,
+		apiBase: apiBase,
+		httpClient: &http.Client{
+			Timeout: 60 * time.Second,
+		},
 	}
 }
 
+func (t *GroqTranscriber) TranscribeData(ctx context.Context, data []byte, filename string) (*TranscriptionResponse, error) {
+	logger.InfoCF("voice", "Starting memory transcription", map[string]any{"filename": filename, "bytes": len(data)})
+
+	var requestBody bytes.Buffer
+	writer := multipart.NewWriter(&requestBody)
+
+	part, err := writer.CreateFormFile("file", filename)
+	if err != nil {
+		logger.ErrorCF("voice", "Failed to create form file", map[string]any{"error": err})
+		return nil, fmt.Errorf("failed to create form file: %w", err)
+	}
+
+	if _, err := io.Copy(part, bytes.NewReader(data)); err != nil {
+		logger.ErrorCF("voice", "Failed to copy file content", map[string]any{"error": err})
+		return nil, fmt.Errorf("failed to copy file content: %w", err)
+	}
+
+	if err = writer.WriteField("model", "whisper-large-v3-turbo"); err != nil {
+		logger.ErrorCF("voice", "Failed to write model field", map[string]any{"error": err})
+		return nil, fmt.Errorf("failed to write model field: %w", err)
+	}
+
+	if err = writer.WriteField("response_format", "json"); err != nil {
+		logger.ErrorCF("voice", "Failed to write response_format field", map[string]any{"error": err})
+		return nil, fmt.Errorf("failed to write response_format field: %w", err)
+	}
+
+	if err = writer.Close(); err != nil {
+		logger.ErrorCF("voice", "Failed to close multipart writer", map[string]any{"error": err})
+		return nil, fmt.Errorf("failed to close multipart writer: %w", err)
+	}
+
+	return t.doRequest(ctx, &requestBody, writer.FormDataContentType(), int64(len(data)))
+}
+
+func (t *GroqTranscriber) Transcribe(ctx context.Context, audioFilePath string) (*TranscriptionResponse, error) {
+	logger.InfoCF("voice", "Starting transcription", map[string]any{"audio_file": audioFilePath})
+
+	audioFile, err := os.Open(audioFilePath)
+	if err != nil {
+		return nil, fmt.Errorf("failed to open audio file: %w", err)
+	}
+	defer audioFile.Close()
+
+	fileInfo, err := audioFile.Stat()
+	if err != nil {
+		return nil, err
+	}
+
+	var requestBody bytes.Buffer
+	writer := multipart.NewWriter(&requestBody)
+	part, err := writer.CreateFormFile("file", filepath.Base(audioFilePath))
+	if err != nil {
+		return nil, err
+	}
+	io.Copy(part, audioFile)
+	writer.WriteField("model", "whisper-large-v3")
+	writer.WriteField("response_format", "json")
+	writer.Close()
+
+	return t.doRequest(ctx, &requestBody, writer.FormDataContentType(), fileInfo.Size())
+}
+
+func (t *GroqTranscriber) doRequest(ctx context.Context, requestBody *bytes.Buffer, contentType string, fileSize int64) (*TranscriptionResponse, error) {
+	url := t.apiBase + "/audio/transcriptions"
+	req, err := http.NewRequestWithContext(ctx, "POST", url, requestBody)
+	if err != nil {
+		logger.ErrorCF("voice", "Failed to create request", map[string]any{"error": err})
+		return nil, fmt.Errorf("failed to create request: %w", err)
+	}
+
+	req.Header.Set("Content-Type", contentType)
+	req.Header.Set("Authorization", "Bearer "+t.apiKey)
+
+	logger.DebugCF("voice", "Sending transcription request to Groq API", map[string]any{
+		"url":                url,
+		"request_size_bytes": requestBody.Len(),
+		"file_size_bytes":    fileSize,
+	})
+
+	resp, err := t.httpClient.Do(req)
+	if err != nil {
+		logger.ErrorCF("voice", "Failed to send request", map[string]any{"error": err})
+		return nil, fmt.Errorf("failed to send request: %w", err)
+	}
+	defer resp.Body.Close()
+
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		logger.ErrorCF("voice", "Failed to read response", map[string]any{"error": err})
+		return nil, fmt.Errorf("failed to read response: %w", err)
+	}
+
+	if resp.StatusCode != http.StatusOK {
+		logger.ErrorCF("voice", "API error", map[string]any{
+			"status_code": resp.StatusCode,
+			"response":    string(body),
+		})
+		return nil, fmt.Errorf("API error (status %d): %s", resp.StatusCode, string(body))
+	}
+
+	logger.DebugCF("voice", "Received response from Groq API", map[string]any{
+		"status_code":         resp.StatusCode,
+		"response_size_bytes": len(body),
+	})
+
+	var result TranscriptionResponse
+	if err := json.Unmarshal(body, &result); err != nil {
+		logger.ErrorCF("voice", "Failed to unmarshal response", map[string]any{"error": err})
+		return nil, fmt.Errorf("failed to unmarshal response: %w", err)
+	}
+
+	logger.InfoCF("voice", "Transcription completed successfully", map[string]any{
+		"text_length":           len(result.Text),
+		"language":              result.Language,
+		"duration_seconds":      result.Duration,
+		"transcription_preview": utils.Truncate(result.Text, 50),
+	})
+
+	return &result, nil
+}
+
+func (t *GroqTranscriber) Name() string {
+	return "groq"
+}
+
 // DetectTranscriber inspects cfg and returns the appropriate Transcriber, or
 // nil if no supported transcription provider is configured.
 func DetectTranscriber(cfg *config.Config) Transcriber {