[FFmpeg-trac] #10535(avfilter:new): astats bit depth, first number not very useful

[FFmpeg]

#10535: astats bit depth, first number not very useful
----------------------------------+---------------------------------------
             Reporter:  danadam   |                     Type:  enhancement
               Status:  new       |                 Priority:  normal
            Component:  avfilter  |                  Version:  git-master
             Keywords:  astats    |               Blocked By:
             Blocking:            |  Reproduced by developer:  0
Analyzed by developer:  0         |
----------------------------------+---------------------------------------
 "Bit depth" line in astats filter output reports two numbers: X/Y

 The meaning of the second number is clear: bits less significant than Y
 are not used.

 The meaning of the first number is: number of bits from Y until the most
 significant bit that toggles to 1.

 Now the thing is, that first number is not very useful. Most of the audio
 has samples with both positive and negative values. With two's complement
 representation, the most significant bit is set to 1 for any negative
 value.

 As a result, for most of the audio the first number will always be the
 same as the second number.

 A more useful value for X would be: max of number of bits from Y until:
 * the most significant bit that toggles to 1, for positive samples,
 * the most significant bit that toggles to 0, plus 1, for negative
 samples.

 This is actually what SoX is reporting in its "stats" effect.

 In attachment I included 2 files: "sine_no_dc.flac" and
 "sine_with_dc.flac".

 "sine_no_dc.flac" is a normal sine, 1 kHz at -25 dBFS with dithering at 14
 bits. This means that top 4 bits and bottom 2 bits are not really used and
 SoX reports it like that:
 {{{
 ]$ sox "sine_no_dc.flac" -n stats
 ...
 Bit-depth      10/14
 }}}
 But because it has samples with negative values, "astats" reports:
 {{{
 ]$ ffmpeg -i "sine_no_dc.flac" -af astats -f null /dev/null
 ...
 [Parsed_astats_0 @ 0x5618db467ac0] Bit depth: 14/14
 }}}

 "sine_with_dc.flac" is similar but it has DC component, so that all
 samples are positive. This time both SoX and ffmpeg report the same:
 {{{
 ]$ sox "sine_with_dc.flac" -n stats
 ...
 Bit-depth       9/14

 ]$ ffmpeg -i "sine_with_dc.flac" -af astats -f null /dev/null
 ...
 [Parsed_astats_0 @ 0x55aa1e5d3b80] Bit depth: 9/14
 }}}

 The files were generated with SoX:
 {{{
 sox -r44.1k -n -b16 "sine_no_dc.flac"   synth 1    sin 1k norm -25 dither
 -p14
 sox -r44.1k -n -b16 "sine_with_dc.flac" synth 1 51 sin 1k norm -25 dither
 -p14
 }}}
-- 
Ticket URL: <https://trac.ffmpeg.org/ticket/10535>
FFmpeg <https://ffmpeg.org>
FFmpeg issue tracker