diff --git a/player/command.c b/player/command.c
index d5a315290e..f8fac69ef7 100644
--- a/player/command.c
+++ b/player/command.c
@@ -7228,6 +7228,7 @@ void mp_option_change_callback(void *ctx, struct m_config_option *co, int flags,
 
     if (opt_ptr == &opts->playback_speed) {
         update_playback_speed(mpctx);
+        reset_av_state(mpctx);
         mp_wakeup_core(mpctx);
     }
 
diff --git a/player/core.h b/player/core.h
index 2fc3c45700..9dc6b8bf90 100644
--- a/player/core.h
+++ b/player/core.h
@@ -636,6 +636,7 @@ void update_osd_msg(struct MPContext *mpctx);
 bool update_subtitles(struct MPContext *mpctx, double video_pts);
 
 // video.c
+void reset_av_state(struct MPContext *mpctx);
 void reset_video_state(struct MPContext *mpctx);
 int init_video_decoder(struct MPContext *mpctx, struct track *track);
 void reinit_video_chain(struct MPContext *mpctx);
diff --git a/player/video.c b/player/video.c
index 70ef18958b..6b63cc7b0e 100644
--- a/player/video.c
+++ b/player/video.c
@@ -45,6 +45,8 @@
 #include "command.h"
 #include "screenshot.h"
 
+#define MIN_PAST_FRAMES 10
+
 enum {
     // update_video() - code also uses: <0 error, 0 eof, >0 progress
     VD_ERROR = -1,
@@ -95,6 +97,17 @@ static void vo_chain_reset_state(struct vo_chain *vo_c)
     vo_c->underrun_signaled = false;
 }
 
+void reset_av_state(struct MPContext *mpctx)
+{
+    mpctx->delay = 0;
+    mpctx->display_sync_drift_dir = 0;
+    mpctx->display_sync_error = 0;
+    mpctx->last_av_difference = 0;
+    mpctx->logged_async_diff = -1;
+    mpctx->num_past_frames = 0;
+    mpctx->total_avsync_change = 0;
+}
+
 void reset_video_state(struct MPContext *mpctx)
 {
     if (mpctx->vo_chain) {
@@ -592,7 +605,9 @@ static void update_avsync_before_frame(struct MPContext *mpctx)
 
     if (mpctx->video_status < STATUS_READY) {
         mpctx->time_frame = 0;
-    } else if (mpctx->display_sync_active || vo->opts->video_sync == VS_NONE) {
+    } else if (mpctx->display_sync_active || vo->opts->video_sync == VS_NONE ||
+               mpctx->num_past_frames <= MIN_PAST_FRAMES)
+    {
         // don't touch the timing
     } else if (mpctx->audio_status == STATUS_PLAYING &&
                mpctx->video_status == STATUS_PLAYING &&
@@ -726,7 +741,7 @@ static double compute_audio_drift(struct MPContext *mpctx, double vsync)
     // audio desync for y. Assume speed didn't change for the frames we're
     // looking at for simplicity. This also should actually use the realtime
     // (minus paused time) for x, but use vsync scheduling points instead.
-    if (mpctx->num_past_frames <= 10)
+    if (mpctx->num_past_frames <= MIN_PAST_FRAMES)
         return NAN;
     int num = mpctx->num_past_frames - 1;
     double sum_x = 0, sum_y = 0, sum_xy = 0, sum_xx = 0;
@@ -831,6 +846,9 @@ static void handle_display_sync_frame(struct MPContext *mpctx,
     if (resample && using_spdif_passthrough(mpctx))
         return;
 
+    if (mpctx->num_past_frames <= MIN_PAST_FRAMES)
+        return;
+
     double vsync = vo_get_vsync_interval(vo) / 1e9;
     if (vsync <= 0)
         return;