OSDN Git Service

Fix timestamps on CC's they are now 100% ready and working. Thanks to j45 and ritsuka...
[handbrake-jp/handbrake-jp-git.git] / libhb / sync.c
index bb94079..eee4708 100644 (file)
@@ -1,13 +1,13 @@
 /* $Id: sync.c,v 1.38 2005/04/14 21:57:58 titer Exp $
 
    This file is part of the HandBrake source code.
-   Homepage: <http://handbrake.m0k.org/>.
+   Homepage: <http://handbrake.fr/>.
    It may be used under the terms of the GNU General Public License. */
 
 #include "hb.h"
-
+#include "hbffmpeg.h"
+#include <stdio.h>
 #include "samplerate.h"
-#include "ffmpeg/avcodec.h"
 
 #ifdef INT64_MIN
 #undef INT64_MIN /* Because it isn't defined correctly in Zeta */
 typedef struct
 {
     hb_audio_t * audio;
-    int64_t      count_frames;
-    
+
+    int64_t      next_start;    /* start time of next output frame */
+    int64_t      next_pts;      /* start time of next input frame */
+    int64_t      first_drop;    /* PTS of first 'went backwards' frame dropped */
+    int          drop_count;    /* count of 'time went backwards' drops */
+
     /* Raw */
     SRC_STATE  * state;
     SRC_DATA     data;
@@ -34,19 +38,28 @@ typedef struct
 struct hb_work_private_s
 {
     hb_job_t * job;
-    int        done;
-
+    int        busy;            // bitmask with one bit for each active input
+                                // (bit 0 = video; 1 = audio 0, 2 = audio 1, ...
+                                // appropriate bit is cleared when input gets
+                                // an eof buf. syncWork returns done when all
+                                // bits are clear.
     /* Video */
-    hb_subtitle_t * subtitle;
     int64_t pts_offset;
-    int64_t pts_offset_old;
-    int64_t count_frames;
-    int64_t count_frames_max;
-    int64_t video_sequence;
+    int64_t next_start;         /* start time of next output frame */
+    int64_t next_pts;           /* start time of next input frame */
+    int64_t first_drop;         /* PTS of first 'went backwards' frame dropped */
+    int drop_count;             /* count of 'time went backwards' drops */
+    int drops;                  /* frames dropped to make a cbr video stream */
+    int dups;                   /* frames duplicated to make a cbr video stream */
+    int video_sequence;
+    int count_frames;
+    int count_frames_max;
+    int chap_mark;              /* to propagate chapter mark across a drop */
     hb_buffer_t * cur; /* The next picture to process */
 
     /* Audio */
     hb_sync_audio_t sync_audio[8];
+    int64_t audio_passthru_slip;
 
     /* Statistics */
     uint64_t st_counts[4];
@@ -58,10 +71,9 @@ struct hb_work_private_s
  * Local prototypes
  **********************************************************************/
 static void InitAudio( hb_work_object_t * w, int i );
-static int  SyncVideo( hb_work_object_t * w );
+static void SyncVideo( hb_work_object_t * w );
 static void SyncAudio( hb_work_object_t * w, int i );
-static int  NeedSilence( hb_work_object_t * w, hb_audio_t * );
-static void InsertSilence( hb_work_object_t * w, int i );
+static void InsertSilence( hb_work_object_t * w, int i, int64_t d );
 static void UpdateState( hb_work_object_t * w );
 
 /***********************************************************************
@@ -82,33 +94,43 @@ int syncInit( hb_work_object_t * w, hb_job_t * job )
 
     pv->job            = job;
     pv->pts_offset     = INT64_MIN;
-    pv->pts_offset_old = INT64_MIN;
-    pv->count_frames   = 0;
 
     /* Calculate how many video frames we are expecting */
-    duration = 0;
-    for( i = job->chapter_start; i <= job->chapter_end; i++ )
+    if (job->pts_to_stop)
+    {
+        duration = job->pts_to_stop + 90000;
+    }
+    else if( job->frame_to_stop )
     {
-        chapter   = hb_list_item( title->list_chapter, i - 1 );
-        duration += chapter->duration;
-    }                                                                           
-    duration += 90000;
+        /* Set the duration to a rough estimate */
+        duration = ( job->frame_to_stop / ( job->vrate / job->vrate_base ) ) * 90000;
+    }
+    else
+    {
+        duration = 0;
+        for( i = job->chapter_start; i <= job->chapter_end; i++ )
+        {
+            chapter   = hb_list_item( title->list_chapter, i - 1 );
+            duration += chapter->duration;
+        }
+        duration += 90000;
         /* 1 second safety so we're sure we won't miss anything */
+    }
     pv->count_frames_max = duration * job->vrate / job->vrate_base / 90000;
 
-    hb_log( "sync: expecting %lld video frames", pv->count_frames_max );
+    hb_log( "sync: expecting %d video frames", pv->count_frames_max );
+    pv->busy |= 1;
 
     /* Initialize libsamplerate for every audio track we have */
-    for( i = 0; i < hb_list_count( title->list_audio ); i++ )
+    if ( ! job->indepth_scan )
     {
-        InitAudio( w, i );
+        for( i = 0; i < hb_list_count( title->list_audio ) && i < 8; i++ )
+        {
+            pv->busy |= ( 1 << (i + 1) );
+            InitAudio( w, i );
+        }
     }
 
-    /* Get subtitle info, if any */
-    pv->subtitle = hb_list_item( title->list_subtitle, 0 );
-
-    pv->video_sequence = 0;
-
     return 0;
 }
 
@@ -122,14 +144,26 @@ void syncClose( hb_work_object_t * w )
     hb_work_private_t * pv = w->private_data;
     hb_job_t          * job   = pv->job;
     hb_title_t        * title = job->title;
-    
+    hb_audio_t        * audio = NULL;
     int i;
 
-    if( pv->cur ) hb_buffer_close( &pv->cur );
+    if( pv->cur )
+    {
+        hb_buffer_close( &pv->cur );
+    }
+
+    hb_log( "sync: got %d frames, %d expected",
+            pv->count_frames, pv->count_frames_max );
+
+    if (pv->drops || pv->dups )
+    {
+        hb_log( "sync: %d frames dropped, %d duplicated", pv->drops, pv->dups );
+    }
 
     for( i = 0; i < hb_list_count( title->list_audio ); i++ )
     {
-        if( job->acodec & HB_ACODEC_AC3 )
+        audio = hb_list_item( title->list_audio, i );
+        if( audio->config.out.codec == HB_ACODEC_AC3 )
         {
             free( pv->sync_audio[i].ac3_buf );
         }
@@ -138,7 +172,7 @@ void syncClose( hb_work_object_t * w )
             src_delete( pv->sync_audio[i].state );
         }
     }
-    
+
     free( pv );
     w->private_data = NULL;
 }
@@ -159,17 +193,16 @@ int syncWork( hb_work_object_t * w, hb_buffer_t ** unused1,
     hb_work_private_t * pv = w->private_data;
     int i;
 
-    /* If we ever got a video frame, handle audio now */
-    if( pv->pts_offset != INT64_MIN )
+    if ( pv->busy & 1 )
+        SyncVideo( w );
+
+    for( i = 0; i < hb_list_count( pv->job->title->list_audio ); i++ )
     {
-        for( i = 0; i < hb_list_count( pv->job->title->list_audio ); i++ )
-        {
+        if ( pv->busy & ( 1 << (i + 1) ) )
             SyncAudio( w, i );
-        }
     }
 
-    /* Handle video */
-    return SyncVideo( w );
+    return ( pv->busy? HB_WORK_OK : HB_WORK_DONE );
 }
 
 hb_work_object_t hb_sync =
@@ -191,7 +224,7 @@ static void InitAudio( hb_work_object_t * w, int i )
     sync        = &pv->sync_audio[i];
     sync->audio = hb_list_item( title->list_audio, i );
 
-    if( job->acodec & HB_ACODEC_AC3 )
+    if( sync->audio->config.out.codec == HB_ACODEC_AC3 )
     {
         /* Have a silent AC-3 frame ready in case we have to fill a
            gap */
@@ -202,11 +235,11 @@ static void InitAudio( hb_work_object_t * w, int i )
         codec = avcodec_find_encoder( CODEC_ID_AC3 );
         c     = avcodec_alloc_context();
 
-        c->bit_rate    = sync->audio->bitrate;
-        c->sample_rate = sync->audio->rate;
-        c->channels    = 2;
+        c->bit_rate    = sync->audio->config.in.bitrate;
+        c->sample_rate = sync->audio->config.in.samplerate;
+        c->channels    = HB_INPUT_CH_LAYOUT_GET_DISCRETE_COUNT( sync->audio->config.in.channel_layout );
 
-        if( avcodec_open( c, codec ) < 0 )
+        if( hb_avcodec_open( c, codec ) < 0 )
         {
             hb_log( "sync: avcodec_open failed" );
             return;
@@ -214,8 +247,8 @@ static void InitAudio( hb_work_object_t * w, int i )
 
         zeros          = calloc( AC3_SAMPLES_PER_FRAME *
                                  sizeof( short ) * c->channels, 1 );
-        sync->ac3_size = sync->audio->bitrate * AC3_SAMPLES_PER_FRAME /
-                             sync->audio->rate / 8;
+        sync->ac3_size = sync->audio->config.in.bitrate * AC3_SAMPLES_PER_FRAME /
+                             sync->audio->config.in.samplerate / 8;
         sync->ac3_buf  = malloc( sync->ac3_size );
 
         if( avcodec_encode_audio( c, sync->ac3_buf, sync->ac3_size,
@@ -223,321 +256,456 @@ static void InitAudio( hb_work_object_t * w, int i )
         {
             hb_log( "sync: avcodec_encode_audio failed" );
         }
-        
+
         free( zeros );
-        avcodec_close( c );
+        hb_avcodec_close( c );
         av_free( c );
     }
     else
     {
         /* Initialize libsamplerate */
         int error;
-        sync->state             = src_new( SRC_LINEAR, HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(sync->audio->amixdown), &error );
+        sync->state             = src_new( SRC_SINC_MEDIUM_QUALITY, HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(sync->audio->config.out.mixdown), &error );
         sync->data.end_of_input = 0;
     }
 }
 
-
-
-#define PTS_DISCONTINUITY_TOLERANCE 90000
-
 /***********************************************************************
  * SyncVideo
  ***********************************************************************
- * 
+ *
  **********************************************************************/
-static int SyncVideo( hb_work_object_t * w )
+static void SyncVideo( hb_work_object_t * w )
 {
     hb_work_private_t * pv = w->private_data;
     hb_buffer_t * cur, * next, * sub = NULL;
     hb_job_t * job = pv->job;
-    int64_t pts_expected;
-    int chap_break;
+    hb_subtitle_t *subtitle;
+    int i;
 
-    if( pv->done )
+    if( !pv->cur && !( pv->cur = hb_fifo_get( job->fifo_raw ) ) )
     {
-        return HB_WORK_DONE;
+        /* We haven't even got a frame yet */
+        return;
     }
-
-    if( hb_thread_has_exited( job->reader ) &&
-        !hb_fifo_size( job->fifo_mpeg2 ) &&
-        !hb_fifo_size( job->fifo_raw ) )
+    cur = pv->cur;
+    if( cur->size == 0 )
     {
-        /* All video data has been processed already, we won't get
-           more */
-        hb_log( "sync: got %lld frames, %lld expected",
-                pv->count_frames, pv->count_frames_max );
-        pv->done = 1;
-        
-        hb_buffer_t * buf_tmp;
+        /* we got an end-of-stream. Feed it downstream & signal that we're done. */
+        hb_fifo_push( job->fifo_sync, hb_buffer_init( 0 ) );
 
-       // Drop an empty buffer into our output to ensure that things
-       // get flushed all the way out.
-        buf_tmp = hb_buffer_init(0); // Empty end buffer
-        hb_fifo_push( job->fifo_sync, buf_tmp );
-        
-        return HB_WORK_DONE;
-    }
+        /*
+         * Push through any subtitle EOFs in case they were not synced through.
+         */
+        for( i = 0; i < hb_list_count( job->list_subtitle ); i++)
+        {
+            subtitle = hb_list_item( job->list_subtitle, i );
+            if( subtitle->dest == PASSTHRUSUB )
+            {
+                hb_fifo_push( subtitle->fifo_out, hb_buffer_init( 0 ) );
+            }
+        }
 
-    if( !pv->cur && !( pv->cur = hb_fifo_get( job->fifo_raw ) ) )
-    {
-        /* We haven't even got a frame yet */
-        return HB_WORK_OK;
+        pv->busy &=~ 1;
+        return;
     }
-    cur = pv->cur;
 
     /* At this point we have a frame to process. Let's check
         1) if we will be able to push into the fifo ahead
         2) if the next frame is there already, since we need it to
-           know whether we'll have to repeat the current frame or not */
+           compute the duration of the current frame*/
     while( !hb_fifo_is_full( job->fifo_sync ) &&
            ( next = hb_fifo_see( job->fifo_raw ) ) )
     {
         hb_buffer_t * buf_tmp;
 
+        if( next->size == 0 )
+        {
+            /* we got an end-of-stream. Feed it downstream & signal that
+             * we're done. Note that this means we drop the final frame of
+             * video (we don't know its duration). On DVDs the final frame
+             * is often strange and dropping it seems to be a good idea. */
+            hb_fifo_push( job->fifo_sync, hb_buffer_init( 0 ) );
+
+            /*
+             * Push through any subtitle EOFs in case they were not synced through.
+             */
+            for( i = 0; i < hb_list_count( job->list_subtitle ); i++)
+            {
+                subtitle = hb_list_item( job->list_subtitle, i );
+                if( subtitle->dest == PASSTHRUSUB )
+                {
+                    hb_fifo_push( subtitle->fifo_out, hb_buffer_init( 0 ) );
+                }
+            }
+            pv->busy &=~ 1;
+            return;
+        }
         if( pv->pts_offset == INT64_MIN )
         {
             /* This is our first frame */
-            hb_log( "sync: first pts is %lld", cur->start );
-            pv->pts_offset = cur->start;
+            pv->pts_offset = 0;
+            if ( cur->start != 0 )
+            {
+                /*
+                 * The first pts from a dvd should always be zero but
+                 * can be non-zero with a transport or program stream since
+                 * we're not guaranteed to start on an IDR frame. If we get
+                 * a non-zero initial PTS extend its duration so it behaves
+                 * as if it started at zero so that our audio timing will
+                 * be in sync.
+                 */
+                hb_log( "sync: first pts is %lld", cur->start );
+                cur->start = 0;
+            }
+        }
+
+        if( cur->new_chap ) {
+            hb_log("sync got new chapter %d", cur->new_chap );
         }
 
         /*
+         * since the first frame is always 0 and the upstream reader code
+         * is taking care of adjusting for pts discontinuities, we just have
+         * to deal with the next frame's start being in the past. This can
+         * happen when the PTS is adjusted after data loss but video frame
+         * reordering causes some frames with the old clock to appear after
+         * the clock change. This creates frames that overlap in time which
+         * looks to us like time going backward. The downstream muxing code
+         * can deal with overlaps of up to a frame time but anything larger
+         * we handle by dropping frames here.
+         */
+        if ( (int64_t)( next->start - cur->start ) <= 0 ||
+             (int64_t)( (cur->start - pv->audio_passthru_slip ) - pv->next_pts ) < 0 )
+        {
+            if ( pv->first_drop == 0 )
+            {
+                pv->first_drop = next->start;
+            }
+            ++pv->drop_count;
+            buf_tmp = hb_fifo_get( job->fifo_raw );
+            if ( buf_tmp->new_chap )
+            {
+                // don't drop a chapter mark when we drop the buffer
+                pv->chap_mark = buf_tmp->new_chap;
+            }
+            hb_buffer_close( &buf_tmp );
+            continue;
+        }
+        if ( pv->first_drop )
+        {
+            hb_log( "sync: video time didn't advance - dropped %d frames "
+                    "(delta %d ms, current %lld, next %lld, dur %d)",
+                    pv->drop_count, (int)( cur->start - pv->first_drop ) / 90,
+                    cur->start, next->start, (int)( next->start - cur->start ) );
+            pv->first_drop = 0;
+            pv->drop_count = 0;
+        }
 
+        /*
          * Track the video sequence number localy so that we can sync the audio
          * to it using the sequence number as well as the PTS.
          */
         pv->video_sequence = cur->sequence;
 
-        /* Check for PTS jumps over 0.5 second */
-        if( next->start < cur->start - PTS_DISCONTINUITY_TOLERANCE ||
-            next->start > cur->start + PTS_DISCONTINUITY_TOLERANCE )
+        /*
+         * Look for a subtitle for this frame.
+         *
+         * If found then it will be tagged onto a video buffer of the correct time and 
+         * sent in to the render pipeline. This only needs to be done for VOBSUBs which
+         * get rendered, other types of subtitles can just sit in their raw_queue until
+         * delt with at muxing.
+         */
+        for( i = 0; i < hb_list_count( job->list_subtitle ); i++)
         {
-           hb_log( "Sync: Video PTS discontinuity (current buffer start=%lld, next buffer start=%lld), trash Video",
-                    cur->start, next->start );
-            
+            subtitle = hb_list_item( job->list_subtitle, i );
+
             /*
-             * Do we need to trash the subtitle, is it from the next->start period
-             * or is it from our old position. If the latter then trash it.
+             * Rewrite timestamps on subtitles that need it (on raw queue).
              */
-            if( pv->subtitle )
+            if( subtitle->source == CC608SUB ||
+                subtitle->source == CC708SUB )
             {
-                while( ( sub = hb_fifo_see( pv->subtitle->fifo_raw ) ) )
+                /*
+                 * Rewrite timestamps on subtitles that came from Closed Captions
+                 * since they are using the MPEG2 timestamps.
+                 */
+                while( ( sub = hb_fifo_see( subtitle->fifo_raw ) ) )
                 {
-                    if( ( sub->start > ( cur->start - PTS_DISCONTINUITY_TOLERANCE ) ) &&
-                        ( sub->start < ( cur->start + PTS_DISCONTINUITY_TOLERANCE ) ) )
+                    /*
+                     * Rewrite the timestamps as and when the video
+                     * (cur->start) reaches the same timestamp as a
+                     * closed caption (sub->start).
+                     *
+                     * What about discontinuity boundaries - not delt
+                     * with here - Van?
+                     *
+                     * Bypass the sync fifo altogether.
+                     */
+                    if( sub->size <= 0 )
                     {
+                        sub = hb_fifo_get( subtitle->fifo_raw );
+                        hb_fifo_push( subtitle->fifo_out, sub );
+                        sub = NULL;
+                        break;
+                    } else {
                         /*
-                         * The subtitle is from our current time region which we are
-                         * jumping from. So trash it as we are about to jump backwards
-                         * or forwards and don't want it blocking the subtitle fifo.
+                         * Sync the subtitles to the incoming video, and use
+                         * the matching converted video timestamp.
+                         *
+                         * Note that it doesn't appear that we need to convert 
+                         * timestamps, I guess that they were already correct,
+                         * so just push them through for rendering.
+                         *
                          */
-                        hb_log("Trashing subtitle 0x%x due to PTS discontinuity", sub);
-                        sub = hb_fifo_get( pv->subtitle->fifo_raw );
-                        hb_buffer_close( &sub );
+                        if( sub->start < cur->start )
+                        {
+                            uint64_t duration;
+                            duration = sub->stop - sub->start;
+                            sub = hb_fifo_get( subtitle->fifo_raw );
+                            hb_fifo_push( subtitle->fifo_out, sub );
+                        } else {
+                            sub = NULL;
+                            break;
+                        }
                     }
                 }
             }
 
-            /* Trash current picture */
-            /* Also, make sure we don't trash a chapter break */
-            chap_break = cur->new_chap;
-            hb_buffer_close( &cur );
-            pv->cur = cur = hb_fifo_get( job->fifo_raw );
-            cur->new_chap |= chap_break; // Don't stomp existing chapter breaks
-
-            /* Calculate new offset */
-            pv->pts_offset_old = pv->pts_offset;
-            pv->pts_offset     = cur->start -
-                pv->count_frames * pv->job->vrate_base / 300;
-
-            pv->video_sequence = cur->sequence;
-            continue;
-        }
-
-        /* Look for a subtitle for this frame */
-        if( pv->subtitle )
-        {
-            hb_buffer_t * sub2;
-            while( ( sub = hb_fifo_see( pv->subtitle->fifo_raw ) ) )
+            if( subtitle->source == VOBSUB ) 
             {
-                /* If two subtitles overlap, make the first one stop
-                   when the second one starts */
-                sub2 = hb_fifo_see2( pv->subtitle->fifo_raw );
-                if( sub2 && sub->stop > sub2->start )
-                    sub->stop = sub2->start;
-
-                // hb_log("0x%x: video seq: %lld  subtitle sequence: %lld", 
-                //       sub, cur->sequence, sub->sequence);
-
-                if( sub->sequence > cur->sequence )
+                hb_buffer_t * sub2;
+                while( ( sub = hb_fifo_see( subtitle->fifo_raw ) ) )
                 {
-                    /*
-                     * The video is behind where we are, so wait until
-                     * it catches up to the same reader point on the
-                     * DVD. Then our PTS should be in the same region
-                     * as the video.
-                     */
-                    sub = NULL;
-                    break;
-                }
-
-                if( sub->stop > cur->start ) {
-                    /*
-                     * The stop time is in the future, so fall through
-                     * and we'll deal with it in the next block of
-                     * code.
-                     */
-                    break;
-                } 
-                else 
-                {
-                    /*
-                     * The stop time is in the past. But is it due to
-                     * it having been played already, or has the PTS
-                     * been reset to 0?
-                     */
-                    if( ( cur->start - sub->stop ) > PTS_DISCONTINUITY_TOLERANCE ) {
+                    if( sub->size == 0 )
+                    {
                         /*
-                         * There is a lot of time between our current
-                         * video and where this subtitle is ending,
-                         * assume that we are about to reset the PTS
-                         * and do not throw away this subtitle.
+                         * EOF, pass it through immediately.
                          */
                         break;
                     }
-                }
 
-                /* 
-                 * The subtitle is older than this picture, trash it 
-                 */
-                sub = hb_fifo_get( pv->subtitle->fifo_raw );
-                hb_buffer_close( &sub );
-            }
-
-            /*
-             * There is a valid subtitle, is it time to display it?
-             */
-            if( sub )
-            {
-                if( sub->stop > sub->start)
-                {
-                    /*
-                     * Normal subtitle which ends after it starts, check to 
-                     * see that the current video is between the start and end.
-                     */
-                    if( cur->start > sub->start &&
-                        cur->start < sub->stop )
+                    /* If two subtitles overlap, make the first one stop
+                       when the second one starts */
+                    sub2 = hb_fifo_see2( subtitle->fifo_raw );
+                    if( sub2 && sub->stop > sub2->start )
+                        sub->stop = sub2->start;
+                    
+                    // hb_log("0x%x: video seq: %lld  subtitle sequence: %lld",
+                    //       sub, cur->sequence, sub->sequence);
+                    
+                    if( sub->sequence > cur->sequence )
                     {
                         /*
-                         * We should be playing this, so leave the
-                         * subtitle in place.
-                         *
-                         * fall through to display
+                         * The video is behind where we are, so wait until
+                         * it catches up to the same reader point on the
+                         * DVD. Then our PTS should be in the same region
+                         * as the video.
                          */
-                    } 
-                    else
-                    {
+                        sub = NULL;
+                        break;
+                    }
+                    
+                    if( sub->stop > cur->start ) {
                         /*
-                         * Defer until the play point is within the subtitle
+                         * The stop time is in the future, so fall through
+                         * and we'll deal with it in the next block of
+                         * code.
                          */
-                        sub = NULL;
+                        break;
                     }
+                    
+                    /*
+                     * The subtitle is older than this picture, trash it
+                     */
+                    sub = hb_fifo_get( subtitle->fifo_raw );
+                    hb_buffer_close( &sub );
                 }
-                else
+                
+                if( sub && sub->size == 0 )
                 {
-                    /*
-                     * The end of the subtitle is less than the start, this is a
-                     * sign of a PTS discontinuity.
+                    /* 
+                     * Continue immediately on subtitle EOF
                      */
-                    if( sub->start > cur->start )
+                    break;
+                }
+
+                /*
+                 * There is a valid subtitle, is it time to display it?
+                 */
+                if( sub )
+                {
+                    if( sub->stop > sub->start)
                     {
                         /*
-                         * we haven't reached the start time yet, or
-                         * we have jumped backwards after having
-                         * already started this subtitle.
+                         * Normal subtitle which ends after it starts, check to
+                         * see that the current video is between the start and end.
                          */
-                        if( cur->start < sub->stop )
+                        if( cur->start > sub->start &&
+                            cur->start < sub->stop )
                         {
                             /*
-                             * We have jumped backwards and so should
-                             * continue displaying this subtitle.
+                             * We should be playing this, so leave the
+                             * subtitle in place.
                              *
-                             * fall through to display.
+                             * fall through to display
                              */
-                        } 
-                        else 
+                            if( ( sub->stop - sub->start ) < ( 3 * 90000 ) )
+                            {
+                                /*
+                                 * Subtitle is on for less than three seconds, extend
+                                 * the time that it is displayed to make it easier
+                                 * to read. Make it 3 seconds or until the next
+                                 * subtitle is displayed.
+                                 *
+                                 * This is in response to Indochine which only
+                                 * displays subs for 1 second - too fast to read.
+                                 */
+                                sub->stop = sub->start + ( 3 * 90000 );
+                                
+                                sub2 = hb_fifo_see2( subtitle->fifo_raw );
+                                
+                                if( sub2 && sub->stop > sub2->start )
+                                {
+                                    sub->stop = sub2->start;
+                                }
+                            }
+                        }
+                        else
                         {
                             /*
                              * Defer until the play point is within the subtitle
                              */
                             sub = NULL;
                         }
-                    } else {
+                    }
+                    else
+                    {
                         /*
-                         * Play this subtitle as the start is greater than our
-                         * video point.
-                         *
-                         * fall through to display/
+                         * The end of the subtitle is less than the start, this is a
+                         * sign of a PTS discontinuity.
                          */
+                        if( sub->start > cur->start )
+                        {
+                            /*
+                             * we haven't reached the start time yet, or
+                             * we have jumped backwards after having
+                             * already started this subtitle.
+                             */
+                            if( cur->start < sub->stop )
+                            {
+                                /*
+                                 * We have jumped backwards and so should
+                                 * continue displaying this subtitle.
+                                 *
+                                 * fall through to display.
+                                 */
+                            }
+                            else
+                            {
+                                /*
+                                 * Defer until the play point is within the subtitle
+                                 */
+                                sub = NULL;
+                            }
+                        } else {
+                            /*
+                             * Play this subtitle as the start is greater than our
+                             * video point.
+                             *
+                             * fall through to display/
+                             */
+                        }
                     }
                 }
             }
-        }
-
-        /* The PTS of the frame we are expecting now */
-        pts_expected = pv->pts_offset +
-            pv->count_frames * pv->job->vrate_base / 300;
-
-        //hb_log("Video expecting PTS %lld, current frame: %lld, next frame: %lld, cf: %lld", 
-        //       pts_expected, cur->start, next->start, pv->count_frames * pv->job->vrate_base / 300 );
+            if( sub )
+            {
+                /*
+                 * Got a sub to display...
+                 */
+                break;
+            }
+        } // end subtitles
 
-        if( cur->start < pts_expected - pv->job->vrate_base / 300 / 2 &&
-            next->start < pts_expected + pv->job->vrate_base / 300 / 2 )
+        /*
+         * Adjust the pts of the current frame so that it's contiguous
+         * with the previous frame. The start time of the current frame
+         * has to be the end time of the previous frame and the stop
+         * time has to be the start of the next frame.  We don't
+         * make any adjustments to the source timestamps other than removing
+         * the clock offsets (which also removes pts discontinuities).
+         * This means we automatically encode at the source's frame rate.
+         * MP2 uses an implicit duration (frames end when the next frame
+         * starts) but more advanced containers like MP4 use an explicit
+         * duration. Since we're looking ahead one frame we set the
+         * explicit stop time from the start time of the next frame.
+         */
+        buf_tmp = cur;
+        pv->cur = cur = hb_fifo_get( job->fifo_raw );
+        pv->next_pts = cur->start;
+        int64_t duration = cur->start - buf_tmp->start;
+        if ( duration <= 0 )
         {
-            /* The current frame is too old but the next one matches,
-               let's trash */
-            /* Also, make sure we don't trash a chapter break */
-            chap_break = cur->new_chap;
-            hb_buffer_close( &cur );
-            pv->cur = cur = hb_fifo_get( job->fifo_raw );
-            cur->new_chap |= chap_break; // Make sure we don't stomp the existing one.
-            
-            continue;
+            hb_log( "sync: invalid video duration %lld, start %lld, next %lld",
+                    duration, buf_tmp->start, next->start );
         }
 
-        if( next->start > pts_expected + 3 * pv->job->vrate_base / 300 / 2 )
-        {
-            /* We'll need the current frame more than one time. Make a
-               copy of it and keep it */
-            buf_tmp = hb_buffer_init( cur->size );
-            memcpy( buf_tmp->data, cur->data, cur->size ); 
-            buf_tmp->sequence = cur->sequence;
-        }
-        else
+        buf_tmp->start = pv->next_start;
+        pv->next_start += duration;
+        buf_tmp->stop = pv->next_start;
+
+        if ( pv->chap_mark )
         {
-            /* The frame has the expected date and won't have to be
-               duplicated, just put it through */
-            buf_tmp = cur;
-            pv->cur = cur = hb_fifo_get( job->fifo_raw );
+            // we have a pending chapter mark from a recent drop - put it on this
+            // buffer (this may make it one frame late but we can't do any better).
+            buf_tmp->new_chap = pv->chap_mark;
+            pv->chap_mark = 0;
         }
-        
-        /* Replace those MPEG-2 dates with our dates */
-        buf_tmp->start = (uint64_t) pv->count_frames *
-            pv->job->vrate_base / 300;
-        buf_tmp->stop  = (uint64_t) ( pv->count_frames + 1 ) *
-            pv->job->vrate_base / 300;
 
         /* If we have a subtitle for this picture, copy it */
         /* FIXME: we should avoid this memcpy */
-        if( sub )
+        if( sub && subtitle && 
+            subtitle->format == PICTURESUB )
         {
-            buf_tmp->sub         = hb_buffer_init( sub->size );
-            buf_tmp->sub->x      = sub->x;
-            buf_tmp->sub->y      = sub->y;
-            buf_tmp->sub->width  = sub->width;
-            buf_tmp->sub->height = sub->height;
-            memcpy( buf_tmp->sub->data, sub->data, sub->size );
+            if( sub->size > 0 )
+            {
+                if( subtitle->dest == RENDERSUB )
+                {
+                    /*
+                     * Tack onto the video buffer for rendering
+                     */
+                    buf_tmp->sub         = hb_buffer_init( sub->size );
+                    buf_tmp->sub->x      = sub->x;
+                    buf_tmp->sub->y      = sub->y;
+                    buf_tmp->sub->width  = sub->width;
+                    buf_tmp->sub->height = sub->height;
+                    memcpy( buf_tmp->sub->data, sub->data, sub->size ); 
+                } else {
+                    /*
+                     * Pass-Through, pop it off of the raw queue, rewrite times and
+                     * make it available to be reencoded.
+                     */
+                    uint64_t sub_duration;
+                    sub = hb_fifo_get( subtitle->fifo_raw );
+                    sub_duration = sub->stop - sub->start;
+                    sub->start = buf_tmp->start;
+                    sub->stop = sub->start + duration;
+                    hb_fifo_push( subtitle->fifo_sync, sub );
+                }
+            } else {
+                /*
+                 * EOF - consume for rendered, else pass through
+                 */
+                if( subtitle->dest == RENDERSUB )
+                {
+                    sub = hb_fifo_get( subtitle->fifo_raw );
+                    hb_buffer_close( &sub );
+                } else {
+                    sub = hb_fifo_get( subtitle->fifo_raw );
+                    hb_fifo_push( subtitle->fifo_out, sub );
+                }
+            }
         }
 
         /* Push the frame to the renderer */
@@ -545,282 +713,255 @@ static int SyncVideo( hb_work_object_t * w )
 
         /* Update UI */
         UpdateState( w );
+        
+        if( job->frame_to_stop && pv->count_frames > job->frame_to_stop )
+        {
+            // Drop an empty buffer into our output to ensure that things
+            // get flushed all the way out.
+            hb_fifo_push( job->fifo_sync, hb_buffer_init( 0 ) );
+            pv->busy &=~ 1;
+            hb_log( "sync: reached %d frames, exiting early (%i busy)",
+                    pv->count_frames, pv->busy );
+            return;
+        }
 
         /* Make sure we won't get more frames then expected */
-        if( pv->count_frames >= pv->count_frames_max )
+        if( pv->count_frames >= pv->count_frames_max * 2)
         {
-            hb_log( "sync: got %lld frames", pv->count_frames );
-            pv->done = 1;
-            
-           // Drop an empty buffer into our output to ensure that things
-           // get flushed all the way out.
-           buf_tmp = hb_buffer_init(0); // Empty end buffer
-           hb_fifo_push( job->fifo_sync, buf_tmp );
-            
-            break;
+            hb_log( "sync: got too many frames (%d), exiting early",
+                    pv->count_frames );
+
+            // Drop an empty buffer into our output to ensure that things
+            // get flushed all the way out.
+            hb_fifo_push( job->fifo_sync, hb_buffer_init( 0 ) );
+            pv->busy &=~ 1;
+            return;
         }
     }
+}
+
+static void OutputAudioFrame( hb_job_t *job, hb_audio_t *audio, hb_buffer_t *buf,
+                              hb_sync_audio_t *sync, hb_fifo_t *fifo, int i )
+{
+    int64_t start = sync->next_start;
+    int64_t duration = buf->stop - buf->start;
+
+    sync->next_pts += duration;
 
-    return HB_WORK_OK;
+    if( audio->config.in.samplerate == audio->config.out.samplerate ||
+        audio->config.out.codec == HB_ACODEC_AC3 ||
+        audio->config.out.codec == HB_ACODEC_DCA )
+    {
+        /*
+         * If we don't have to do sample rate conversion or this audio is 
+         * pass-thru just send the input buffer downstream after adjusting
+         * its timestamps to make the output stream continuous.
+         */
+    }
+    else
+    {
+        /* Not pass-thru - do sample rate conversion */
+        int count_in, count_out;
+        hb_buffer_t * buf_raw = buf;
+        int channel_count = HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(audio->config.out.mixdown) *
+                            sizeof( float );
+
+        count_in  = buf_raw->size / channel_count;
+        /*
+         * When using stupid rates like 44.1 there will always be some
+         * truncation error. E.g., a 1536 sample AC3 frame will turn into a
+         * 1536*44.1/48.0 = 1411.2 sample frame. If we just truncate the .2
+         * the error will build up over time and eventually the audio will
+         * substantially lag the video. libsamplerate will keep track of the
+         * fractional sample & give it to us when appropriate if we give it
+         * an extra sample of space in the output buffer.
+         */
+        count_out = ( duration * audio->config.out.samplerate ) / 90000 + 1;
+
+        sync->data.input_frames = count_in;
+        sync->data.output_frames = count_out;
+        sync->data.src_ratio = (double)audio->config.out.samplerate /
+                               (double)audio->config.in.samplerate;
+
+        buf = hb_buffer_init( count_out * channel_count );
+        sync->data.data_in  = (float *) buf_raw->data;
+        sync->data.data_out = (float *) buf->data;
+        if( src_process( sync->state, &sync->data ) )
+        {
+            /* XXX If this happens, we're screwed */
+            hb_log( "sync: audio %d src_process failed", i );
+        }
+        hb_buffer_close( &buf_raw );
+
+        buf->size = sync->data.output_frames_gen * channel_count;
+        duration = ( sync->data.output_frames_gen * 90000 ) /
+                   audio->config.out.samplerate;
+    }
+    buf->frametype = HB_FRAME_AUDIO;
+    buf->start = start;
+    buf->stop  = start + duration;
+    sync->next_start = start + duration;
+    hb_fifo_push( fifo, buf );
 }
 
 /***********************************************************************
  * SyncAudio
  ***********************************************************************
- * 
+ *
  **********************************************************************/
 static void SyncAudio( hb_work_object_t * w, int i )
 {
     hb_work_private_t * pv = w->private_data;
-    hb_job_t        * job;
-    hb_audio_t      * audio;
+    hb_job_t        * job = pv->job;
+    hb_sync_audio_t * sync = &pv->sync_audio[i];
+    hb_audio_t      * audio = sync->audio;
     hb_buffer_t     * buf;
-    hb_sync_audio_t * sync;
-
     hb_fifo_t       * fifo;
-    int               rate;
+    int64_t start;
 
-    int64_t           pts_expected;
-    int64_t           start;
-
-    job    = pv->job;
-    sync   = &pv->sync_audio[i];
-    audio  = sync->audio;
-
-    if( job->acodec & HB_ACODEC_AC3 )
+    if( audio->config.out.codec == HB_ACODEC_AC3 ||
+        audio->config.out.codec == HB_ACODEC_DCA )
     {
-        fifo = audio->fifo_out;
-        rate = audio->rate;
+        fifo = audio->priv.fifo_out;
     }
     else
     {
-        fifo = audio->fifo_sync;
-        rate = job->arate;
+        fifo = audio->priv.fifo_sync;
     }
 
-    while( !hb_fifo_is_full( fifo ) &&
-           ( buf = hb_fifo_see( audio->fifo_raw ) ) )
+    while( !hb_fifo_is_full( fifo ) && ( buf = hb_fifo_see( audio->priv.fifo_raw ) ) )
     {
-        /* The PTS of the samples we are expecting now */
-        pts_expected = pv->pts_offset + sync->count_frames * 90000 / rate;
-
-        // hb_log("Video Sequence: %lld, Audio Sequence: %lld", pv->video_sequence, buf->sequence);
-        /*
-         * Using the same logic as the Video have we crossed a VOB
-         * boundary as detected by the expected PTS and the PTS of our
-         * audio being out by more than the tolerance value.
-         */
-        if( ( buf->start > pts_expected + PTS_DISCONTINUITY_TOLERANCE ||
-              buf->start < pts_expected - PTS_DISCONTINUITY_TOLERANCE ) &&
-            pv->pts_offset_old > INT64_MIN )
+        start = buf->start - pv->audio_passthru_slip;
+        /* if the next buffer is an eof send it downstream */
+        if ( buf->size <= 0 )
         {
-           /*
-             * Useful debug, but too verbose for normal use.
-             */
-            hb_log("Sync: Audio discontinuity (sequence: vid %lld aud %lld) (pts %lld < %lld < %lld)",
-                   pv->video_sequence, buf->sequence,
-                  pts_expected - PTS_DISCONTINUITY_TOLERANCE, buf->start,
-                  pts_expected + PTS_DISCONTINUITY_TOLERANCE );
-           
-            /* There has been a PTS discontinuity, and this frame might
-               be from before the discontinuity*/
-            pts_expected = pv->pts_offset_old + sync->count_frames *
-                90000 / rate;
-
-            /*
-             * Is the audio from a valid period given the previous
-             * Video PTS. I.e. has there just been a video PTS
-             * discontinuity and this audio belongs to the vdeo from
-             * before?
-             */
-            if( buf->start > pts_expected + PTS_DISCONTINUITY_TOLERANCE ||
-                buf->start < pts_expected - PTS_DISCONTINUITY_TOLERANCE )
-            {
-                /*
-                 * It's outside of our tolerance for where the video
-                 * is now, and it's outside of the tolerance for
-                 * where we have been in the case of a VOB change.
-                 * Try and reconverge regardless. so continue on to
-                 * our convergence code below which will kick in as
-                 * it will be more than 100ms out.
-                 * 
-                 * Note that trashing the Audio could make things
-                 * worse if the Audio is in front because we will end
-                 * up diverging even more. We need to hold on to the
-                 * audio until the video catches up.
-                 */
-                hb_log("Sync: Audio is way out of sync, attempt to reconverge from current video PTS");
-               
-                /*
-                 * It wasn't from the old place, so we must be from
-                 * the new, but just too far out. So attempt to
-                 * reconverge by resetting the point we want to be to
-                 * where we are currently wanting to be.
-                 */
-               pts_expected = pv->pts_offset + sync->count_frames * 90000 / rate;
-                start = pts_expected - pv->pts_offset;
-           } else {
-                 /* Use the older offset */
-                start = pts_expected - pv->pts_offset_old;
-           }
+            buf = hb_fifo_get( audio->priv.fifo_raw );
+            hb_fifo_push( fifo, buf );
+            pv->busy &=~ (1 << (i + 1) );
+            return;
         }
-        else
+        if( job->frame_to_stop && pv->count_frames >= job->frame_to_stop )
         {
-            start = pts_expected - pv->pts_offset;
+            hb_fifo_push( fifo, hb_buffer_init(0) );
+            pv->busy &=~ (1 << (i + 1) );
+            return;
         }
-
-        /* Tolerance: 100 ms */
-        if( buf->start < pts_expected - 9000 )
+        if ( (int64_t)( start - sync->next_pts ) < 0 )
         {
-           /* Audio is behind the Video, trash it, can't use it now. */
-           hb_log( "Sync: Audio PTS (%lld) < Video PTS (%lld) by greater than 100ms, trashing audio to reconverge",
-                     buf->start, pts_expected);
-            buf = hb_fifo_get( audio->fifo_raw );
-            hb_buffer_close( &buf );
-            continue;
+            // audio time went backwards.
+            // If our output clock is more than a half frame ahead of the
+            // input clock drop this frame to move closer to sync.
+            // Otherwise drop frames until the input clock matches the output clock.
+            if ( sync->first_drop || sync->next_start - start > 90*15 )
+            {
+                // Discard data that's in the past.
+                if ( sync->first_drop == 0 )
+                {
+                    sync->first_drop = sync->next_pts;
+                }
+                ++sync->drop_count;
+                buf = hb_fifo_get( audio->priv.fifo_raw );
+                hb_buffer_close( &buf );
+                continue;
+            }
+            sync->next_pts = start;
         }
-        else if( buf->start > pts_expected + 9000 )
+        if ( sync->first_drop )
         {
-            /* Audio is ahead of the Video, insert silence until we catch up*/
-           hb_log("Sync: Audio PTS (%lld) >  Video PTS (%lld) by greater than 100ms insert silence until reconverged", buf->start, pts_expected);
-            InsertSilence( w, i );
-            continue;
+            // we were dropping old data but input buf time is now current
+            hb_log( "sync: audio %d time went backwards %d ms, dropped %d frames "
+                    "(next %lld, current %lld)", i,
+                    (int)( sync->next_pts - sync->first_drop ) / 90,
+                    sync->drop_count, sync->first_drop, sync->next_pts );
+            sync->first_drop = 0;
+            sync->drop_count = 0;
+            sync->next_pts = start;
         }
-
-        if( job->acodec & HB_ACODEC_AC3 )
+        if ( start - sync->next_pts >= (90 * 70) )
         {
-            buf        = hb_fifo_get( audio->fifo_raw );
-            buf->start = start;
-            buf->stop  = start + 90000 * AC3_SAMPLES_PER_FRAME / rate;
-
-            sync->count_frames += AC3_SAMPLES_PER_FRAME;
-        }
-        else
-        {
-            hb_buffer_t * buf_raw = hb_fifo_get( audio->fifo_raw );
-
-            int count_in, count_out;
-
-            count_in  = buf_raw->size / HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(audio->amixdown) / sizeof( float );
-            count_out = ( buf_raw->stop - buf_raw->start ) * job->arate / 90000;
-            if( buf->start < pts_expected - 1500 )
-                count_out--;
-            else if( buf->start > pts_expected + 1500 )
-                count_out++;
-
-            sync->data.data_in      = (float *) buf_raw->data;
-            sync->data.input_frames = count_in;
-            sync->data.output_frames = count_out;
-
-            sync->data.src_ratio = (double) sync->data.output_frames /
-                                   (double) sync->data.input_frames;
-
-            buf = hb_buffer_init( sync->data.output_frames * HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(audio->amixdown) *
-                                  sizeof( float ) );
-            sync->data.data_out = (float *) buf->data;
-            if( src_process( sync->state, &sync->data ) )
+            if ( start - sync->next_pts > (90000LL * 60) )
             {
-                /* XXX If this happens, we're screwed */
-                hb_log( "sync: src_process failed" );
+                // there's a gap of more than a minute between the last
+                // frame and this. assume we got a corrupted timestamp
+                // and just drop the next buf.
+                hb_log( "sync: %d minute time gap in audio %d - dropping buf"
+                        "  start %lld, next %lld",
+                        (int)((start - sync->next_pts) / (90000*60)),
+                        i, start, sync->next_pts );
+                buf = hb_fifo_get( audio->priv.fifo_raw );
+                hb_buffer_close( &buf );
+                continue;
             }
-            hb_buffer_close( &buf_raw );
-
-            buf->size = sync->data.output_frames_gen * HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(audio->amixdown) * sizeof( float );
-
-            /* Set dates for resampled data */
-            buf->start = start;
-            buf->stop  = start + sync->data.output_frames_gen *
-                            90000 / job->arate;
-
-            sync->count_frames += sync->data.output_frames_gen;
+            /*
+             * there's a gap of at least 70ms between the last
+             * frame we processed & the next. Fill it with silence.
+             * Or in the case of DCA, skip some frames from the
+             * other streams.
+             */
+            if( sync->audio->config.out.codec == HB_ACODEC_DCA )
+            {
+                hb_log( "sync: audio gap %d ms. Skipping frames. Audio %d"
+                        "  start %lld, next %lld",
+                        (int)((start - sync->next_pts) / 90),
+                        i, start, sync->next_pts );
+                pv->audio_passthru_slip += (start - sync->next_pts);
+                return;
+            }
+            hb_log( "sync: adding %d ms of silence to audio %d"
+                    "  start %lld, next %lld",
+                    (int)((start - sync->next_pts) / 90),
+                    i, start, sync->next_pts );
+            InsertSilence( w, i, start - sync->next_pts );
+            return;
         }
 
-        buf->frametype = HB_FRAME_AUDIO;
-        hb_fifo_push( fifo, buf );
-    }
-
-    if( NeedSilence( w, audio ) )
-    {
-        InsertSilence( w, i );
-    }
-}
-
-static int NeedSilence( hb_work_object_t * w, hb_audio_t * audio )
-{
-    hb_work_private_t * pv = w->private_data;
-    hb_job_t * job = pv->job;
-
-    if( hb_fifo_size( audio->fifo_in ) ||
-        hb_fifo_size( audio->fifo_raw ) ||
-        hb_fifo_size( audio->fifo_sync ) ||
-        hb_fifo_size( audio->fifo_out ) )
-    {
-        /* We have some audio, we are fine */
-        return 0;
-    }
-
-    /* No audio left in fifos */
-
-    if( hb_thread_has_exited( job->reader ) )
-    {
-        /* We might miss some audio to complete encoding and muxing
-           the video track */
-       hb_log("Reader has exited early, inserting silence.");
-        return 1;
-    }
-
-    if( hb_fifo_is_full( job->fifo_mpeg2 ) &&
-        hb_fifo_is_full( job->fifo_raw ) &&
-        hb_fifo_is_full( job->fifo_sync ) &&
-        hb_fifo_is_full( job->fifo_render ) &&
-        hb_fifo_is_full( job->fifo_mpeg4 ) )
-    {
-        /* Too much video and no audio, oh-oh */
-       hb_log("Still got some video - and nothing in the audio fifo, insert silence");
-        return 1;
+        /*
+         * When we get here we've taken care of all the dups and gaps in the
+         * audio stream and are ready to inject the next input frame into
+         * the output stream.
+         */
+        buf = hb_fifo_get( audio->priv.fifo_raw );
+        OutputAudioFrame( job, audio, buf, sync, fifo, i );
     }
-
-    return 0;
 }
 
-static void InsertSilence( hb_work_object_t * w, int i )
+static void InsertSilence( hb_work_object_t * w, int i, int64_t duration )
 {
     hb_work_private_t * pv = w->private_data;
-    hb_job_t        * job;
-    hb_sync_audio_t * sync;
-    hb_buffer_t     * buf;
-
-    job    = pv->job;
-    sync   = &pv->sync_audio[i];
-
-    if( job->acodec & HB_ACODEC_AC3 )
-    {
-        buf        = hb_buffer_init( sync->ac3_size );
-        buf->start = sync->count_frames * 90000 / sync->audio->rate;
-        buf->stop  = buf->start + 90000 * AC3_SAMPLES_PER_FRAME /
-                     sync->audio->rate;
-        memcpy( buf->data, sync->ac3_buf, buf->size );
-
-        hb_log( "sync: adding a silent AC-3 frame for track %x",
-                sync->audio->id );
-        hb_fifo_push( sync->audio->fifo_out, buf );
-
-        sync->count_frames += AC3_SAMPLES_PER_FRAME;
-
-    }
-    else
+    hb_job_t        *job = pv->job;
+    hb_sync_audio_t *sync = &pv->sync_audio[i];
+    hb_buffer_t     *buf;
+    hb_fifo_t       *fifo;
+
+    // to keep pass-thru and regular audio in sync we generate silence in
+    // AC3 frame-sized units. If the silence duration isn't an integer multiple
+    // of the AC3 frame duration we will truncate or round up depending on
+    // which minimizes the timing error.
+    const int frame_dur = ( 90000 * AC3_SAMPLES_PER_FRAME ) /
+                          sync->audio->config.in.samplerate;
+    int frame_count = ( duration + (frame_dur >> 1) ) / frame_dur;
+
+    while ( --frame_count >= 0 )
     {
-        buf        = hb_buffer_init( HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(sync->audio->amixdown) * job->arate / 20 *
-                                     sizeof( float ) );
-        buf->start = sync->count_frames * 90000 / job->arate;
-        buf->stop  = buf->start + 90000 / 20;
-        memset( buf->data, 0, buf->size );
-
-        hb_log( "sync: adding 50 ms of silence for track %x",
-                sync->audio->id );
-        hb_fifo_push( sync->audio->fifo_sync, buf );
-
-        sync->count_frames += job->arate / 20;
+        if( sync->audio->config.out.codec == HB_ACODEC_AC3 )
+        {
+            buf        = hb_buffer_init( sync->ac3_size );
+            buf->start = sync->next_pts;
+            buf->stop  = buf->start + frame_dur;
+            memcpy( buf->data, sync->ac3_buf, buf->size );
+            fifo = sync->audio->priv.fifo_out;
+        }
+        else
+        {
+            buf = hb_buffer_init( AC3_SAMPLES_PER_FRAME * sizeof( float ) *
+                                     HB_AMIXDOWN_GET_DISCRETE_CHANNEL_COUNT(
+                                         sync->audio->config.out.mixdown) );
+            buf->start = sync->next_pts;
+            buf->stop  = buf->start + frame_dur;
+            memset( buf->data, 0, buf->size );
+            fifo = sync->audio->priv.fifo_sync;
+        }
+        OutputAudioFrame( job, sync->audio, buf, sync, fifo, i );
     }
 }
 
@@ -843,14 +984,14 @@ static void UpdateState( hb_work_object_t * w )
                  3 * sizeof( uint64_t ) );
         pv->st_dates[3]  = hb_get_date();
         pv->st_counts[3] = pv->count_frames;
-    } 
+    }
 
 #define p state.param.working
     state.state = HB_STATE_WORKING;
     p.progress  = (float) pv->count_frames / (float) pv->count_frames_max;
     if( p.progress > 1.0 )
     {
-        p.progress = 1.0; 
+        p.progress = 1.0;
     }
     p.rate_cur   = 1000.0 *
         (float) ( pv->st_counts[3] - pv->st_counts[0] ) /