calls: treat empty opus packets as DTX, not decode errors

calls/audio/xmpp_to_tg.go

@@ -50,6 +50,7 @@ type XmppToTg struct {
 	cShipSilence  atomic.Int64 // of those, silence frames (underrun/priming)
 	cPlcFrames    atomic.Int64 // PLC 10ms frames synthesised over gaps
 	cGapEvents    atomic.Int64 // ticks where a seq gap was skipped
+	cDtxFrames    atomic.Int64 // empty (DTX) packets concealed as comfort noise
 	cSendErrs     atomic.Int64 // SendMicrophonePCM errors
 
 	capture *captureWriter
@@ -57,7 +58,7 @@ type XmppToTg struct {
 
 // counter snapshot for computing per-interval deltas in the stats log
 type xmppToTgSnap struct {
-	rtpIn, rtpSkippedPT, shipTotal, shipSilence, plcFrames, gapEvents, sendErrs int64
+	rtpIn, rtpSkippedPT, shipTotal, shipSilence, plcFrames, gapEvents, dtxFrames, sendErrs int64
 }
 
 func (h *XmppToTg) snap() xmppToTgSnap {
@@ -68,6 +69,7 @@ func (h *XmppToTg) snap() xmppToTgSnap {
 		shipSilence:  h.cShipSilence.Load(),
 		plcFrames:    h.cPlcFrames.Load(),
 		gapEvents:    h.cGapEvents.Load(),
+		dtxFrames:    h.cDtxFrames.Load(),
 		sendErrs:     h.cSendErrs.Load(),
 	}
 }
@@ -273,6 +275,7 @@ func (h *XmppToTg) logStats(elapsed, sinceStart time.Duration, ticks, stalls, ma
 		"feed_drift_ms":   feedDriftMs,
 		"plc_frames":      cur.plcFrames - prev.plcFrames,
 		"gap_events":      cur.gapEvents - prev.gapEvents,
+		"dtx_frames":      cur.dtxFrames - prev.dtxFrames,
 		"send_errs":       cur.sendErrs - prev.sendErrs,
 		"depth":           ps.Depth,
 		"depth_ms":        ps.Depth * OpusFrameMs,
@@ -314,6 +317,19 @@ func (h *XmppToTg) processTick(st *tickState) {
 			st.pending10ms = appendChunks10ms(st.pending10ms, plcOut)
 			h.cPlcFrames.Add(int64(len(st.pending10ms) - before))
 		}
+		if len(pkt.Payload) == 0 {
+			// empty payload = opus DTX/comfort-noise frame during silence, not
+			// a decode error. Conceal for the last frame's duration (libopus
+			// synthesises comfort noise) rather than emitting hard silence.
+			h.cDtxFrames.Add(1)
+			plcOut := h.pcmBuf[:st.lastDecodedSamples*Channels]
+			if err := h.dec.DecodePLC(plcOut); err != nil {
+				h.cShipSilence.Add(1)
+				h.sendToNtg(h.silence)
+				return
+			}
+			st.pending10ms = appendChunks10ms(st.pending10ms, plcOut)
+		} else {
 			n, err := h.dec.Decode(pkt.Payload, h.pcmBuf)
 			if err != nil {
 				h.log.WithError(err).Warn("opus decode failed")
@@ -324,6 +340,7 @@ func (h *XmppToTg) processTick(st *tickState) {
 			st.lastDecodedSamples = n
 			st.pending10ms = appendChunks10ms(st.pending10ms, h.pcmBuf[:n*Channels])
 		}
+	}
 	chunk := st.pending10ms[0]
 	st.pending10ms = st.pending10ms[1:]
 	h.sendToNtg(chunk)

calls/audio/xmpp_to_tg_test.go

@@ -119,6 +119,35 @@ func TestProcessTickDecodesAndSplits(t *testing.T) {
 	}
 }
 
+func TestProcessTickConcealsDtxEmptyPacket(t *testing.T) {
+	h, n := newTestXmppToTg(t)
+	enc, err := NewEncoder()
+	if err != nil {
+		t.Fatalf("NewEncoder: %v", err)
+	}
+	// Prime with a real packet, then an empty (DTX) packet the peer sends
+	// during silence.
+	h.playout.Push(rtpWithSeq(200, encodeOpusFrame(t, enc)))
+	h.playout.Push(rtpWithSeq(201, nil))
+
+	st := newTickState()
+	// real packet -> 2 chunks (ticks 1-2), DTX packet -> concealed (ticks 3+).
+	for i := 0; i < 4; i++ {
+		h.processTick(st)
+	}
+
+	if got := h.cDtxFrames.Load(); got != 1 {
+		t.Errorf("dtx frames = %d, want 1 (empty packet must be concealed, not errored)", got)
+	}
+	// DTX must not fall through to the silence path, and every tick ships a frame.
+	if got := h.cShipSilence.Load(); got != 0 {
+		t.Errorf("ship_silence = %d, want 0 (DTX conceals rather than emitting silence)", got)
+	}
+	if got := n.sentCount(); got != 4 {
+		t.Errorf("shipped %d frames, want 4", got)
+	}
+}
+
 func TestProcessTickRunsPLCBeforeGapRecovery(t *testing.T) {
 	h, _ := newTestXmppToTg(t)
 	enc, err := NewEncoder()