[FFmpeg-devel] [PATCH] matroskadec: Improve TTA duration calculation

[James Almer]

Calculate the duration as accurately as possible to improve decoding of samples 
where the last frame is smaller than the rest.

Example:
Take lossless-audio/luckynight-partial.tak from the FATE suit and convert it 
to TTA muxed into matroska:

ffmpeg -i $(SAMPLES)/lossless-audio/luckynight-partial.tak -c:a tta lucky.mka

The framemd5 output for lucky.mka without this patch:

0,          0,          0,    46080,   184320, 7c3751ddd571d2903c3cf0ab4b3e3d0a
0,      46080,      46080,    46080,   184320, 6b70c782ba1da3f933fde2daa4f96b73
0,      92160,      92160,    46080,   184320, dcf70d89c54b9a4f0b302d4ec4fb302d
0,     138240,     138240,    46080,   184320, 48713ca38b388d2ea4abf5b86ed1226f
0,     184320,     184320,    46080,   184320, 12188a23648e7ebfb07cd6fe9197b2ea
0,     230400,     230400,    46080,   184320, 49653ab8186a5d4a044ed284671a26e0
0,     276480,     276480,    46080,   184320, 5e82c6a7fe58c7ea612c03a0a2927dd4
0,     322560,     322560,    46080,   184320, 83dc449dbd9eab5f2e8ad2b4403d6a21
0,     368640,     368640,    46080,   184320, bdd6b92c23d30978d4e802d305b0fc49

With this patch:

0,          0,          0,    46080,   184320, 7c3751ddd571d2903c3cf0ab4b3e3d0a
0,      46080,      46080,    46080,   184320, 6b70c782ba1da3f933fde2daa4f96b73
0,      92160,      92160,    46080,   184320, dcf70d89c54b9a4f0b302d4ec4fb302d
0,     138240,     138240,    46080,   184320, 48713ca38b388d2ea4abf5b86ed1226f
0,     184320,     184320,    46080,   184320, 12188a23648e7ebfb07cd6fe9197b2ea
0,     230400,     230400,    46080,   184320, 49653ab8186a5d4a044ed284671a26e0
0,     276480,     276480,    46080,   184320, 5e82c6a7fe58c7ea612c03a0a2927dd4
0,     322560,     322560,    46080,   184320, 83dc449dbd9eab5f2e8ad2b4403d6a21
0,     368640,     368640,    46080,   184320, bdd6b92c23d30978d4e802d305b0fc49
0,     414720,     414720,     4230,    16920, b50b440c5bbcecb8e9fbece643447593

The duration without this patch was calculated as 418950000000, which is bigger 
than uint32_t and as such stored as 2338172288.
With this patch the duration is correctly calculated as 418950.

Signed-off-by: James Almer <jamrial at gmail.com>
---
 libavformat/matroskadec.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/libavformat/matroskadec.c b/libavformat/matroskadec.c
index d52877c..b78553e 100644
--- a/libavformat/matroskadec.c
+++ b/libavformat/matroskadec.c
@@ -1738,7 +1738,7 @@ static int matroska_read_header(AVFormatContext *s)
             avio_wl16(&b, track->audio.channels);
             avio_wl16(&b, track->audio.bitdepth);
             avio_wl32(&b, track->audio.out_samplerate);
-            avio_wl32(&b, matroska->ctx->duration * track->audio.out_samplerate);
+            avio_wl32(&b, av_rescale((matroska->duration * matroska->time_scale), track->audio.out_samplerate, AV_TIME_BASE * 1000));
         } else if (codec_id == AV_CODEC_ID_RV10 || codec_id == AV_CODEC_ID_RV20 ||
                    codec_id == AV_CODEC_ID_RV30 || codec_id == AV_CODEC_ID_RV40) {
             extradata_offset = 26;
-- 
1.8.1.5