[FFmpeg-devel] calculation of probe score

[Dave Rice]

Hi Carl,

> On Sep 13, 2016, at 2:57 PM, Carl Eugen Hoyos <ceffmpeg at gmail.com> wrote:
> 
> 2016-09-12 19:47 GMT+02:00 Monique Lassere <lasserem at gmail.com>:
>> Trying to dig as deeply as possible into ffprobe's probe score
>> and how it is calculated.
> 
> Is there a use case that could be interesting to us?

I think the use case if for when an archive accepts a large diverse collection for preservation to access probe_score per file. A lower probe_score may indicate that the file is broken, misnamed, mis-categorized, which could mean that creation of derivatives or long-term handling may need closer review.

In a related conversation about probe score I tested 2000+ random files and made this histogram: rate of occurrence, format name and probe_score.

149 "matroska,webm",100
672 "mov,mp4,m4a,3gp,3g2,mj2",100
  1 aac,1
  1 aac,25
  1 adp,25
 15 aiff,100
  4 asf,100
  8 avi,100
919 caf,100
 45 dv,75
  2 ea,100
  1 flac,1
  1 flac,13
  4 flac,50
  8 flv,100
  3 mp3,25
 90 mp3,51
  9 mpeg,26
  8 mpeg,52
  1 mpegts,50
  1 mpegvideo,12
 13 mpegvideo,51
 11 mxf,100
  1 ogg,100
  2 rm,100
  1 swf,100
  5 swf,26
 92 wav,99

So for instance perhaps worth a closer look at the mpegvideo with score of 12 or the flac file with score of 1(!), before moving the file into long term storage.

> Or a bug you want to tell us about?
> 
> [...]
> 
>> 2. Secondly, does the probe score relate to the container format only?
> 
> As Michael said, yes.
> Note that some containers (like H.264 Annex B or ADTS) correspond
> to a codec, so for containers that "allow" random codecs (like mpeg
> streams) the same probe functions that allow detecting formats can
> be used to detect codecs.

[...]

Best Regards,
Dave Rice