Selur's Little Message Board

Hello,

I have some PAL SD videos captured from digital camera using Windows Media Player over FireWire. They are Type-2 AVIs (the audio stream is stored twice - once interleaved in the video, and once separately, where the interleaved stream is disabled). Here's a sample of ffprobe:


For some reason, the interleaved stream is at 48000 Hz (wrong!), but the real stream is correct at 32000 Hz. This shouldn't be a problem because the wrong stream is disabled anyway. When I encode with Hybrid using "passthrough all", I expect the streams to be preserved. However, it seems like instead, the second one gets resampled to 48000 Hz:


As you would expect, this resampling completely messes the audio when played back.

Out of curiosity, why does Hybrid extract the audio streams and them merge them back in? This sounds like an error-prone process.

Actually, even if you fix the "passthrough" issue, so both streams keep their sample rate, I think that would still break my case because when encoding the final video, Hybrid won't set the second stream as default/the first one as disabled. Would it be possible to just add audio "auto add (last)" the same way there is "auto add (first)"? This will be a nice trick for all Type-2 AVIs. I was planning to work around this issue by using the mkvmerge custom option "--audio-tracks 1" and select the second stream, but that fails since Hybrid already uses "--no-audio" due to the extraction mechanism above.
The reason why I can't just manually make these selections is because I am running an automated pipeline with zero user interaction.

My system:
- Windows 11
- Hybrid 2022.07.01.1

Thank you!

Quote:Out of curiosity, why does Hybrid extract the audio streams and them merge them back in? This sounds like an error-prone process.

Hybrid does process audio and video separately since it offers way more varity in how to deal with the audio and video.
(normally works fine, but I haven't had to work with 32Hz audio for 10+years )

Quote:Would it be possible to just add audio "auto add (last)" the same way there is "auto add (first)"?

Possible in theory, sure. But sorry this won't come soon. (Haven't changed anything in the general auto audio routines for ages, would need some time to re-read how everything is intertwined)
(side note: Hybrid does allow filtering by language, if the audio is tagged accordingly, which your streams doe not seem to be)

If you can share a small sample of such a source I can try to reproduce and fix the sample rate issue.

Quote:They are Type-2 AVIs (the audio stream is stored twice - once interleaved in the video, and once separately, where the interleaved stream is disabled).

Out of curiousity: Who does that and why?

Cu Selur

Quote:Out of curiousity: Who does that and why?

Type-1 and Type-2 AVI dates to the times when computers were severely underpowered. Here is a very good explanation by Microsoft themselves:

Quote:DV cameras produce interleaved audio-video; each frame of video also contains the audio information. If you save DV data to an AVI file, you have a choice:

- Store the interleaved data as one stream in the AVI file. This is known as a type-1 file.
- Split the interleaved data into separate audio and video streams. This is known as a type-2 file.

For video capture, where maximum throughput is crucial, it is better to use a type-1 file, because type-2 files carry redundant audio data. (The video stream still has the audio data. The audio is simply hidden by labeling the stream as video.) Also, writing a type-2 file requires some additional processor time to split the interleaved stream.

On the other hand, type-1 files are less efficient for real-time editing. The application must extract the audio from the interleaved stream, make the edits, and interleave the data again. Also, the type-1 format is not compatible with Microsoft® Video for Windows® (VFW). DirectShow can handle both types of files.

Source: https://docs.microsoft.com/en-us/windows...-avi-files

Quote:If you can share a small sample of such a source I can try to reproduce and fix the sample rate issue.

A sample incoming to your PMs!

Thanks

Okay, the problem seems to be that MediaInfo reports only one audio stream:


FFmpeg/FFprobe/FFplayer report:
MPlayer reports:
First time mediainfo reports less audio streams than libav/ffmpeg -> looking into it.

Cu Selur

Okay,. reading https://forum.videohelp.com/threads/3747...io-streams and testing that with your stream, I that FFmpeg still only copies the first audio stream.

Seems like ffmpeg's (default?) behavior is to copy the first audio track only. I can get it to copy both audio streams by using:
The sample rate is preserved. I can also force the first stream to be skipped like this:
Sadly, in both cases the "correct" audio stream is silent, apart from a high-pitched buzz at the start...

Okay, I know what happens.
MediaInfo reports a sample rate of 48khz for the first audio stream and nothing for the second. Hybrid then assumes that the audio sample rate is 48kHz, which is above what is reported by ffmpeg, and thus Hybrid uses the 48kHz.
I changed that for testing, but like you when using:
to extract the first
and
two extract the second audio, the second audio is silent.

Not sure I should keep my workaround for this case or .avi files. Especially when the second audio when extracted should be the same as the first.

Cu Selur

I'm actually not too sure anymore whether my videos are really Type-2 (one interleaved audio stream, one separate). I tried the tool dvdate (https://paulglagla.com/en/dvdate-2/), and it identified the file as Type-1. When I converted it to Type-2 using dvdate, it identifies like this:


Then I converted the Type-2 back to Type-1 using dvdate, and it looks like this:


Both files produced by dvdate work wonders using ffmpeg and Hybrid. I think in my video's case and Type-1 in general, the second audio stream is actually not a real audio stream, but just the timecode and metadata for the interleaved audio stream, and this is why extracting it while dropping the interleaved stream causes high pitched buzz and no sound. I suspect what Hybrid and ffmpeg need to do, and what VLC Player does, is use the metadata from the second stream and apply it on the content of the first/interleaved stream. Not sure how feasible this is, or whether it's worth the effort at all, since it sounds like a super edge case.

Not feasable for Hybrid, unless there's an option in FFmpeg to do this.

Cu Selur

I agree. I think the main issue here is MediaInfo's and ffmpeg's handling of Type-1 AVIs. MediaInfo reports one audio stream (the real/interleaved stream), while the latter reports two streams (the interleaved stream and the timecode/metadata stream). Both seem unable to apply the timecode/metadata stream's sample rate onto the real/interleaved stream. Only VLC player appears to do that. Honestly though, this is an extreme edge case, since in almost all other videos the metadata of both streams matches anyway. I worked around this issue by converting my problematic videos back and forth using dvdate, which properly applies the correct sample rate to both audio streams. Then, I simply do "auto add (first)" in Hybrid, and I am done.

Fun discovery: Even with the proper Type-1 AVI produced by dvdate, the second audio stream is silent. This confirms my theory that it does not contain any audio content, but merely metadata.