Bookmarks vs. Resume - Bug or feature?

[PSBlake]

It's been a long time since I've posted anything, but here goes:

How exactly does the resume function differ from the bookmark function?  I ask because, in both the ipod Nano and iriver H10/20gb builds, as far back as I've been using rockbox, resume-from-shutdown and resume-from-bookmark behave differently.  Not enough to actually impede functionality or anything, but enough to be noticeable, and enough to make me wonder if the difference is intentional.

Specifically, when resuming from shutdown on a VBR mp3 file, resume will pick up a few seconds after the point where the unit was actually shut down.  The specific amount of time varies depending on the average bitrate of the file in question: Lower bitrates result in a larger jump.  Resuming from bookmark, however, always leads to the exact place at which the bookmark was set.

I have only noticed this behavior with VBR mp3s, although I have not tested it with other VBR file formats.  It does not affect CBR files, insofar as I have noticed.  It has been present in rockbox for as long as I have been aware of rockbox, certainly since the first H10/20gb builds.  It is less noticable in high bitrate files, and almost undetectable at bitrates approaching 160kbps average.  However, I use VBR (rather irrationally) in lower bitrates for audiobooks, and the jump can sometimes result in lost sentences.

[saratoga]

Specifically, when resuming from shutdown on a VBR mp3 file, resume will pick up a few seconds after the point where the unit was actually shut down.  The specific amount of time varies depending on the average bitrate of the file in question: Lower bitrates result in a larger jump.

I just tried this, and couldn't reproduce.  Seems to start from exactly the same point (to within a drum beat anyway) for me using LAME V2 and 128k CBR on a sansa. 

[karashata]

[...] using LAME V2 and 128k CBR on a sansa. 

The OP did say it only seemed to happen on VBR mp3's...

[saratoga]

[...] using LAME V2 and 128k CBR on a sansa. 

The OP did say it only seemed to happen on VBR mp3's...

V2 is VBR.

[karashata]

Oh, my bad then.

[PSBlake]

As a test, I used a text-to-speech program and created a file which consists of nothing more than a single voice counting from 1 to 1000.  Since the text-to-speech program outputs the wav files as mono, 22khz, I re-encoded the file in lame using the following options:

-s 22 -m m -q 0 --vbr-new

The resulting file hovers around 75kbps, and clocks in at around 20 minutes long.  I loaded it onto my H10, and allowed it to count to "two hundred five" (the last syllable  I was able to hear in the fadeout was actually "Five") before shutting the unit down.

When I powered the unit back up and pressed play, the unit resumed at "eight, two hundred nine, two hundred ten..."

So, the jump wasn't big, but it was definitely there.  I generally encode machine read data at a lower bitrate (which results in larger jumps), but due to the EULA of the computer voice I usually use, I couldn't legally distribute the test file (see below).

To test bookmarks, I used the same file.  I paused the file at "two hundred thirty-seven" (actually hearing the word "seven" in the fadeout), created a bookmark, power cycled, then resumed using the bookmark.  The voice picked back up at "...n two hundred thirty-eight...".  Apparently, the bookmark caught just before the end of the "n" sound.  So, no gap with a bookmark.

Mind you, I have only encountered this behavior on the H10 20gb and the ipod Nano 1st gen.  I was unable to test on my Nano, as it has taken to strange behavior which, among other things, completely prevents rockbox from booting.

*VBR Sample file removed.  It was almost certainly legal, but since I can't find an EULA for Microsoft Sam with terms regarding distributing recordings made of the voice, I apparently don't have the right to do so.

[Llorean]

Remember, this forum adheres by whether it's legal in terms of "what the laws and licenses say" not "whether it's actionable."

If the license allows you to redistribute the output, it's fine, but you don't need all that jabber about whether it's actionable. Independent of the morality of the issue, we respect copyright as we expect ours to be respected, so it's what the license says that matters, not what you think about it.

[PSBlake]

Hence the bit about "it doesn't say I can't."  The rest was just in case anyone wanted to muddy the waters with armchair lawyering.  Laws do not enable: They codify restrictions.  Until something is restricted, it is allowed.  There is not currently a law or EULA restriction which disallows distribution of an audio recording of the Microsoft Sam voice counting from 1 to 1000.

Since this was created using freely available tools, and tools which Microsoft did not see fit to restrict the output of, this file is legally free and clear.  I'm just paranoid that there might be people lurking here who would feel the need to pull out some legal boilerplate about distributing modified code, or derivative works, or some other nonsense that doesn't apply.

[saratoga]

Perhaps its an issue with the rockbox resampler.  Are you getting it for 44.1khz audio?

[Llorean]

Most text to speech engines actually do have a license that explicitly states whether you can or cannot redistribute the output of the engine.

[PSBlake]

Perhaps its an issue with the rockbox resampler.  Are you getting it for 44.1khz audio?

Yes, I do encounter this problem with higher frequency files, including some CD-rips I've done of audiobooks.  However, I cannot legally share such samples to test.  I suppose I could upsample the existing test file, but it doesn't really seem necessary.