diff --git a/daemon/codec.c b/daemon/codec.c
index 335d2838b..fb4c8fd43 100644
--- a/daemon/codec.c
+++ b/daemon/codec.c
@@ -184,12 +184,13 @@ reset:
 }
 
 struct codec_handler *codec_handler_make_playback(struct rtp_payload_type *src_pt,
-		struct rtp_payload_type *dst_pt)
+		struct rtp_payload_type *dst_pt, unsigned long last_ts)
 {
 	struct codec_handler *handler = __handler_new(src_pt);
 	handler->dest_pt = *dst_pt;
 	handler->func = handler_func_playback;
 	handler->ssrc_handler = (void *) __ssrc_handler_transcode_new(handler);
+	handler->ssrc_handler->first_ts = last_ts;
 	while (handler->ssrc_handler->first_ts == 0)
 		handler->ssrc_handler->first_ts = random();
 	handler->ssrc_handler->rtp_mark = 1;
@@ -712,6 +713,7 @@ static void __output_rtp(struct media_packet *mp, struct codec_ssrc_handler *ch,
 	payload_tracker_add(&ssrc_out->tracker, handler->dest_pt.payload_type);
 	p->free_func = free;
 	p->source = handler;
+	p->rtp = rh;
 
 	// this packet is dynamically allocated, so we're able to schedule it.
 	// determine scheduled time to send
diff --git a/daemon/media_player.c b/daemon/media_player.c
index c427e15cb..9b89cdf43 100644
--- a/daemon/media_player.c
+++ b/daemon/media_player.c
@@ -250,7 +250,14 @@ found:
 	ilog(LOG_DEBUG, "Output codec for media playback is " STR_FORMAT,
 			STR_FMT(&dst_pt->encoding_with_params));
 
-	mp->handler = codec_handler_make_playback(&src_pt, dst_pt);
+	// if we played anything before, scale our sync TS according to the time
+	// that has passed
+	if (mp->sync_ts_tv.tv_sec) {
+		long long ts_diff_us = timeval_diff(&rtpe_now, &mp->sync_ts_tv);
+		mp->sync_ts += ts_diff_us * dst_pt->clock_rate / 1000000 / dst_pt->codec_def->clockrate_mult;
+	}
+
+	mp->handler = codec_handler_make_playback(&src_pt, dst_pt, mp->sync_ts);
 	if (!mp->handler)
 		return -1;
 
@@ -326,6 +333,16 @@ static void media_player_read_packet(struct media_player *mp) {
 	// update our global "now" timestamp
 	gettimeofday(&rtpe_now, NULL);
 
+	// keep track of RTP timestamps and real clock. look at the last packet we received
+	// and update our sync TS.
+	if (packet.packets_out.head) {
+		struct codec_packet *p = packet.packets_out.head->data;
+		if (p->rtp) {
+			mp->sync_ts = ntohl(p->rtp->timestamp);
+			mp->sync_ts_tv = p->to_send;
+		}
+	}
+
 	mutex_lock(&mp->sink->out_lock);
 	if (media_socket_dequeue(&packet, mp->sink))
 		ilog(LOG_ERR, "Error sending playback media to RTP sink");
diff --git a/include/codec.h b/include/codec.h
index 6f9039696..add8949e3 100644
--- a/include/codec.h
+++ b/include/codec.h
@@ -16,6 +16,7 @@ struct media_packet;
 struct ssrc_hash;
 struct sdp_ng_flags;
 struct codec_ssrc_handler;
+struct rtp_header;
 
 
 typedef int codec_handler_func(struct codec_handler *, struct media_packet *);
@@ -36,6 +37,7 @@ struct codec_handler {
 
 struct codec_packet {
 	str s;
+	struct rtp_header *rtp;
 	void *source; // opaque
 	void (*free_func)(void *);
 	struct timeval to_send;
@@ -45,7 +47,7 @@ struct codec_packet {
 struct codec_handler *codec_handler_get(struct call_media *, int payload_type);
 void codec_handlers_free(struct call_media *);
 struct codec_handler *codec_handler_make_playback(struct rtp_payload_type *src_pt,
-		struct rtp_payload_type *dst_pt);
+		struct rtp_payload_type *dst_pt, unsigned long ts);
 void codec_handler_free(struct codec_handler *handler);
 
 void codec_add_raw_packet(struct media_packet *mp);
diff --git a/include/media_player.h b/include/media_player.h
index 8546447e7..57a1934dc 100644
--- a/include/media_player.h
+++ b/include/media_player.h
@@ -38,6 +38,8 @@ struct media_player {
 	struct codec_handler *handler;
 	struct ssrc_ctx *ssrc_out;
 	unsigned long seq;
+	unsigned long sync_ts;
+	struct timeval sync_ts_tv;
 
 	AVIOContext *avioctx;
 	str *blob;