Use Haali Media Splitter for .ts playback?

[mianpe]

Hello,

Long time GB-PVR user here finally looking into making the switch to NEXTPVR. Got it all setup with new hardware and things are working, but I see laggy performance with Fast Forward and Rewind while watching recordings (ATSC OTA .ts from a Hauppage WinTV tuner). Generally after hitting FF, the audio immediately skips forward and plays (this is good), but the video takes 1-2 seconds to start playing. It's a bit annoying since my 5 year old HTPC setup with GP-PVR works so nicely.

So I got looking at my old setup and I was using Haali Media Splitter. Sure enough, after installing that and loading the .ts file in Graphedit, it works a treat. Snappy and fast. However, it looks like NEXTPVR will never select the Haali splitter, instead choosing "NPVR TS Reader2" and "MPEG-2 Demultiplexer".

So am I out of luck, or is there a way to get NEXTPVR to use the same filters as graphedit for .ts files?

Thanks!
-Mike

[whurlston]

NextPVR will always use "NPVR TS Reader2" and "MPEG-2 Demultiplexer" for .ts files.

[sub]

Sorry, when it comes to .ts file it always use the Microsoft MPEG-2 Demultiplexer.

[mianpe]

Ok, bummer. Thanks for the quick reply though. Any thoughts on why it is so laggy? Do other people experience this? I experimented with converting the .ts file to a .mpg file and that did improve the performance but produced some weird "slow-mo" effect for a couple of seconds after each FF or RW. Would love to hear other ATSC OTA users experiences.

Thanks,
-Mike

[sub]

Ok, bummer. Thanks for the quick reply though. Any thoughts on why it is so laggy?

Do you mean as described in the first post? ie, when skipping you get audio immediately, but don't get video for a second or two?

This can happen depending on how your broadcaster is encoding/muxing the video, and where exactly skip lands. ie, video decoders will typically start showing a picture when it gets an I-Frame. How often these happen come around varies depending on video encoding GOP size. It's usually somewhere around once a second, but some push it out to nearly two seconds.

If you happen to end up skipping to just before an I-Frame, then you'll usually get video straight way. Unfortunately NextPVR cant just always seek to just before an I-Frame, because there is some other factors, like how the closely audio/video packets are muxed in the stream.

My seeking on our NZ Freeview MPEG2 files here are really good. I'll usually get video pretty much straight way when skipping.

[mianpe]

Yes, that is what I meant. I've been playing around with it some more and I find that if I pause playback before I do the fast forward, then the picture updates immediately. This makes it feel more responsive, although it then takes a couple of seconds to start again once I resume play. You may be right about the I-Frames and GOP size, but I do know that the Haali splitter handles these files with ease. I'm assuming there is good reason for forcing the Microsoft Demultiplexer, just wish it performed better for me...

Would still love to hear from anybody that has slow .ts file seeking with NextPVR and what if anything they've done to work around it.

[whurlston]

Have you tested using a different video decoder?

[mianpe]

Yes, but only a few:
AMD Video Decoder
MPC MPEG-2 Video Decoder
Microsoft DTV-DVD Video Decoder

The Microsoft DTV-DVD Video Decoder seems to work the best, and in fact that is the one that gets picked up as default by ZoomPlayer (along with the Haali splitter) and it performs great on these files. (An interesting tidbit is that windows media player picks up the Microsoft Demux and suffers from the same issue as NextPVR.)

Thanks for the suggestion, I'll try a few more decoders today. Maybe it's a combination thing and I'll find a good match that works better. My gut feeling is that it is some kind of buffering issue since the delay seems to occur at the start of playback from a new position in the file.

[mianpe]

Well, I've been having fun playing around with different decoders, but with no significant difference.

I also tried to dig a little deeper into how all this works, and I did find something interesting. I found this pass-through filter for DirectShow which you can insert into GraphEdit and see some interesting timestamps while the graph is running. I placed this filter in between the "NPVR TS Reader2" and the "MPEG-2 Demultiplexer" and also between the demux and the "Microsoft DTV-DVD Video Decoder" and then ran the graph and did some seeking through the file.

Here's an snippet from the logs around a seek event (number on left is time in msec):

--First, between TSReader2 and Demux:

23080 : 24f0a90 Sample 169200 bytes untimed
23231 : 24f0a90 Sample 169200 bytes untimed
23379 : 24f0a90 Sample 169200 bytes untimed
23481 : 24f0a90 Sample 169200 bytes untimed
23608 : 24f0a90 Sample 169200 bytes untimed
23731 : 24f0a90 Sample 169200 bytes untimed
23829 : 24f0a90 Sample 169200 bytes untimed
23958 : 24f0a90 Sample 169200 bytes untimed
23999 : 24f0a90 BeginFlush
24003 : 24f0a90 EndFlush
24004 : NewSegment: 517000000..4611686018427387903 at 1x
24004 : GetState 1
24018 : 24f0a90 Sample 169200 bytesD untimed
24018 : 24f0a90 Run STO 24011ms ST 4ms
24020 : 24f0a90 Sample 169200 bytes untimed
24021 : 24f0a90 Sample 169200 bytes untimed
24024 : 24f0a90 Sample 169200 bytes untimed
24026 : 24f0a90 Sample 169200 bytes untimed
24028 : 24f0a90 Sample 169200 bytes untimed
24029 : 24f0a90 Sample 169200 bytes untimed
24031 : 24f0a90 Sample 169200 bytes untimed
24032 : 24f0a90 Sample 169200 bytes untimed
24033 : 24f0a90 Sample 169200 bytes untimed
24034 : 24f0a90 Sample 169200 bytes untimed
24035 : 24f0a90 Sample 169200 bytes untimed
24036 : 24f0a90 Sample 169200 bytes untimed
24037 : 24f0a90 Sample 169200 bytes untimed
24038 : 24f0a90 Sample 169200 bytes untimed
24041 : 24f0a90 Sample 169200 bytes untimed
25479 : 24f0a90 Sample 169200 bytes untimed
25576 : 24f0a90 Sample 169200 bytes untimed
25708 : 24f0a90 Sample 169200 bytes untimed
25827 : 24f0a90 Sample 169200 bytes untimed
25941 : 24f0a90 Sample 169200 bytes untimed
26076 : 24f0a90 Sample 169200 bytes untimed
26161 : 24f0a90 Sample 169200 bytes untimed
26279 : 24f0a90 Sample 169200 bytes untimed

--Second, between demux and decoder:

23972 : 24f2c20 Sample 8192 bytes untimed
23972 : 24f2c20 Sample 8192 bytes untimed
23972 : 24f2c20 Sample 8192 bytes untimed
23973 : 24f2c20 Sample 8192 bytes untimed
23976 : 24f2c20 Sample 7205 bytes untimed
23976 : 24f2c20 Sample 8192 bytes time 3660.5555 to 3660.5556 latency 36.6140
23976 : 24f2c20 Sample 4122 bytes untimed
23977 : 24f2c20 Sample 8192 bytes time 3677.2444 to 3677.2445 latency 53.1436
23979 : 24f2c20 Sample 3386 bytes untimed
23979 : 24f2c20 Sample 8192 bytes time 3743.9778 to 3743.9779 latency 117.3078
23989 : 24f2c20 Sample 8192 bytes untimed
23992 : 24f2c20 Sample 8192 bytes untimed
23997 : 24f2c20 BeginFlush
24001 : 24f2c20 EndFlush
24016 : 24f2c20 Run STO 24011ms ST 4ms
24018 : 24f2c20 Sample 8192 bytesD time 1101.3725 to 1101.3726 latency 1094.6863
24019 : 24f2c20 Sample 1546 bytes untimed
24019 : 24f2c20 Sample 8192 bytes time 1118.614 to 1118.615 latency 1111.2008
24019 : 24f2c20 Sample 74 bytes untimed
24019 : 24f2c20 Sample 8192 bytes time 1184.7948 to 1184.7949 latency 1177.7746
24019 : 24f2c20 Sample 8192 bytes untimed
24020 : 24f2c20 Sample 8192 bytes untimed
24020 : 24f2c20 Sample 5397 bytes untimed
24020 : 24f2c20 Sample 8192 bytes time 1151.4281 to 1151.4282 latency 1143.2556
24020 : 24f2c20 Sample 1362 bytes untimed
24020 : 24f2c20 Sample 8192 bytes time 1168.1059 to 1168.1060 latency 1159.7792
24020 : 24f2c20 Sample 1178 bytes untimed
24020 : 24f2c20 Sample 8192 bytes time 1234.8392 to 1234.8393 latency 1226.3652
24021 : 24f2c20 Sample 8192 bytes untimed
24021 : 24f2c20 Sample 8192 bytes untimed
24021 : 24f2c20 Sample 4845 bytes untimed
24021 : 24f2c20 Sample 8192 bytes time 1201.4725 to 1201.4726 latency 1191.9111
24022 : 24f2c20 Sample 258 bytes untimed
24022 : 24f2c20 Sample 7530 bytes time 1218.1614 to 1218.1615 latency 1208.1988
24022 : 24f2c20 Sample 8192 bytes time 1284.8948 to 1284.8949 latency 1274.8561
24022 : 24f2c20 Sample 8192 bytes untimed
24025 : 24f2c20 Sample 8192 bytes untimed
24027 : 24f2c20 Sample 2821 bytes untimed
24028 : 24f2c20 Sample 7898 bytes time 1251.5281 to 1251.5282 latency 1235.7479
24028 : 24f2c20 Sample 7898 bytes time 1268.2059 to 1268.2060 latency 1252.3847
24028 : 24f2c20 Sample 8192 bytes time 1334.9392 to 1334.9393 latency 1319.789
24031 : 24f2c20 Sample 8192 bytes untimed
24031 : 24f2c20 Sample 8192 bytes untimed
24031 : 24f2c20 Sample 8192 bytes untimed
24031 : 24f2c20 Sample 1069 bytes untimed
24031 : 24f2c20 Sample 8192 bytes time 1301.5725 to 1301.5726 latency 1282.2899
24031 : 24f2c20 Sample 810 bytes untimed
24031 : 24f2c20 Sample 8192 bytes time 1318.2614 to 1318.2615 latency 1298.8210
24033 : 24f2c20 Sample 1730 bytes untimed
24034 : 24f2c20 Sample 8192 bytes time 1384.9948 to 1384.9949 latency 1362.5931
24034 : 24f2c20 Sample 8192 bytes untimed
24034 : 24f2c20 Sample 8192 bytes untimed
24034 : 24f2c20 Sample 8192 bytes untimed
24034 : 24f2c20 Sample 8192 bytes untimed
24035 : 24f2c20 Sample 8192 bytes untimed
24035 : 24f2c20 Sample 8192 bytes untimed
24035 : 24f2c20 Sample 8192 bytes untimed
24035 : 24f2c20 Sample 8192 bytes untimed
24035 : 24f2c20 Sample 8192 bytes untimed
24035 : 24f2c20 Sample 8192 bytes untimed
24035 : 24f2c20 Sample 8192 bytes untimed
24038 : 24f2c20 Sample 6557 bytes untimed
24038 : 24f2c20 Sample 8192 bytes time 1351.6281 to 1351.6282 latency 1325.5366
24038 : 24f2c20 Sample 5042 bytes untimed
24038 : 24f2c20 Sample 8192 bytes time 1368.3059 to 1368.3060 latency 1341.8183
24039 : 24f2c20 Sample 4122 bytes untimed
24039 : 24f2c20 Sample 8192 bytes time 1435.392 to 1435.393 latency 1407.4956
24039 : 24f2c20 Sample 8192 bytes untimed
24039 : 24f2c20 Sample 8192 bytes untimed
24039 : 24f2c20 Sample 8192 bytes untimed
24039 : 24f2c20 Sample 1989 bytes untimed
24039 : 24f2c20 Sample 8192 bytes time 1401.6725 to 1401.6726 latency 1373.9727
24040 : 24f2c20 Sample 5042 bytes untimed
24040 : 24f2c20 Sample 8192 bytes time 1418.3614 to 1418.3615 latency 1390.4969
24041 : 24f2c20 Sample 626 bytes untimed
24041 : 24f2c20 Sample 8192 bytes time 1485.948 to 1485.949 latency 1455.8953
25388 : 24f2c20 Sample 8192 bytes untimed
25389 : 24f2c20 Sample 8192 bytes untimed
25405 : 24f2c20 Sample 4109 bytes untimed
25408 : 24f2c20 Sample 8192 bytes time 1451.7281 to 1451.7282 latency 55.6797
25422 : 24f2c20 Sample 1914 bytes untimed
25422 : 24f2c20 Sample 8192 bytes time 1468.4059 to 1468.4060 latency 58.4684
25422 : 24f2c20 Sample 3018 bytes untimed
25425 : 24f2c20 Sample 8192 bytes time 1535.1392 to 1535.1393 latency 122.3533

As you can see there is a large latency on the sample from 24018 to 24041, and then no update until 25388 which is 1.347 seconds later.

Anyway, don't know if this is interesting to anybody else. It would be cool if somebody else tried this experiment and reported back their numbers. Also, any suggestions on a better way to debug this problem would be greatly appreciated.

Thanks,
-Mike