Split video outside the key frames?

[KeyMs92]

Hi,

This may well be an obvious question, but just to be sure: Is there any way I can split video's outside of the key frames while maintaining the original video quality?
I can only split each 5 seconds, which is very frustrating
Thanks in advance!

Format                           : Flash Video
File size                        : 439 MiB
Duration                         : 2h 1mn
Overall bit rate                 : 504 Kbps
_presetname                      : Custom
_creationdate                    : Fri Sep 16 13:57:57 2011
_videodevice                     : Microsoft DV Camera and VCR
_framerate                       : 25.000
_width                           : 512.000
_height                          : 288.000
_videocodecid                    : VP62
_videodatarate                   : 400.000
_videokeyframe_frequency         : 5.000
_audiodevice                     : Microsoft DV Camera and VCR
_audiosamplerate                 : 44100.000
_audiochannels                   : 1.000
_audioinputvolume                : 75.000
_audiocodecid                    : .mp3
_audiodatarate                   : 96.000

Video
Format                           : VP6
Codec ID                         : 4
Duration                         : 2h 1mn
Bit rate                         : 380 Kbps
Width                            : 512 pixels
Height                           : 288 pixels
Display aspect ratio             : 16:9
Frame rate mode                  : Variable
Bit depth                        : 8 bits
Stream size                      : 331 MiB (75%)

Audio
Format                           : MPEG Audio
Format version                   : Version 1
Format profile                   : Layer 3
Codec ID                         : 2
Codec ID/Hint                    : MP3
Duration                         : 2h 1mn
Bit rate mode                    : Constant
Bit rate                         : 96.0 Kbps
Channel(s)                       : 1 channel
Sampling rate                    : 44.1 KHz
Compression mode                 : Lossy
Stream size                      : 83.6 MiB (19%)

[Baldrick]

Nope. But some tools can just reencode near the split and keep the rest in original quality. What is your source?

[KeyMs92]

Right, that would be quite useful. File info is above, I got the video from a BBC stream recording. Can you name a few of those tools?

[KeyMs92]

Alright, I've found that e.g. Avidemux and VirtualDub support this feature (sometimes called Smart Rendering). However, in both cases it seems impossible to encode to VP6 video. The closest I'm able to get is FLV1/Sorenson Spark. I'm afraid the VP6 encoder might be commercial.. Any ideas?

[Baldrick]

I don't think any editor supports smart rendering with vp6 video.

[KeyMs92]

Does that mean I've no choice but to convert the whole video to a different format (when I want to accurately edit)?

[jagabo]

You could always do the "smart" part yourself. If you can find an non-reencoding editor that only cuts on key frames you can cut your video into segments. Re-encode the segments you need to cut. Finally join all the segments together. For example, if you need to cut off a part of the first GOP, split the file into two segments, the first GOP and the rest. Cut and reencode the first GOP. Then join that new video with the second segment.

[KeyMs92]

Yes of course, but I haven't found any encoder tool that works with VP6. I've installed the official "VP6 6.4.2.0" codec", but I have no idea how to make it work VirtualDub or any other tool; nothing shows up in the compression codecs lists.

EDIT: Got it! Needed the VP6VFW codec.

[KeyMs92]

Ok, I've experimented a bit and all seems to go well except that the video is partly upside down.
FYI, here's what settings I'm using for the VP6 compression (VP62 Heightened Sharpness). Note that I used 2 passes (of which the image is the second).

As for the VirtualDub settings: I used fast compression, although full processing gave exactly the same file. Also, as it turns out smart rendering does work for (On2) VP6; the encoding is notably faster and the file size differs from having it disabled. What's most interesting though is that the video flips from upside down to normal exactly at the point of the first keyframe, while with smart rendering off it stays normal. Now that tells me that smart rendering is working but is also causing the problem!

[KeyMs92]

Alright, I think I've found the definitive method (if there is one).

1. Using Ultra Video Joiner, extract the approximate part that must be edited (VirtualDub crashes opening the original recording).
2. Open the file in VirtualDub and take note of the video delay (caused by Ultra Video Joiner). Use this value as extra margin at the end of the video. Else the audio is cut off to early.
3. Open the file and extract the to-be-compressed part, i.e. the starting frame until the key frame, using Video Full Processing Mode (to get the filter on), the vertical flip filter (solves the flipping issue) and with Smart Rendering Enabled (to get the codec to match exactly with part two.
4. Extract the rest, starting at the key frame of part one, using Direct Stream Copy.
5. Merge the two parts.
6. Using Audacity, get the audio track from the original output of Ultra Video Joiner, extract the right part and import as audio track for the merged parts and apply the delay.
7. Export to avi and done

The flipping issue remains weird because information is scarce. What is clear to me, is that it is caused by changing the FourCC values in the container. It works like a switch: As soon as you change the value to any other value other than the original, the video flips. As the uncompressed part is exported with a FourCC of VP6F and the encoded part with VP62 you can see the problem arise.

The only real annoyance I still have is that VirtualDub leaves 5ms of silence at the beginning of each file, no matter how low you set the preload block in the settings. This forces me to use Audacity to avoid gaps.
Anyway, thought I would share this (dunno why :P). If you have any suggestions let me know!

[jagabo]

Preload doesn't effect when the audio starts playing. It's how much of the audio appears in the file before the first video frame. The audio still doesn't play until the flagged skew has passed.

[KeyMs92]

Ah I see, that makes sense. I'll probably have to edit it then, to get the transition absolutely smooth.

[KeyMs92]

After having had a look at it again, I noticed the silence is actually caused by using full processing (PCM). If you use stream copy (mp3) there's no silence. Then, if you convert to wav again using Audacity there's no silence either. For some unknown reason VirtualDub adds this 5ms silence at the beginning...

[jagabo]

Were you reencoding when you got the 5ms of silence at the start of a clip? Many audio encoders do that. I've see it with MP3 encoders, for example.

[KeyMs92]

Nope. That's the weird thing, normally this happens when you encode the audio.

EDIT: I also tried the Fraunhofer MPEG Layer-3 Audio Decoder. It results in exactly the same file. Perhaps it's not really the decoding process but rather the way in which the audio (wav) and video track are compiled into the avi file.