2. The _level_ is often extremely low - especially for some old (say 1960-1999) video clips. (Not all, by any means - but often enough to be noticeable.) By low, I mean I have to boost them by ×4, or even ×8 or occasionally ×16, to get the peaks above 50% full scale. (I only use
powers of 2 to avoid distortion.) Is this something YouTube are
imposing? Is audio level adjustment difficult on some common piece of
video capture hardware/software? I even came across one recently where
the uploader _said_ something like "this is quiet, you may have to
adjust" in the notes, so s/he knew about it. This does seem odd.
3. (This is the one that finally prompted me to post.) Far more often
than not - I'd say over 90% of tracks - there's a visible (I can no
longer hear that high) tone around 15½ kHz. I presume in the majority of cases, it's timebase - 15625 for "PAL" (yes I know, but YKWIM), 15750
for NTSC; even where it's not from an actual video source, I presume it
has picked it up somewhere in the processing, e. g. from a computer monitor/graphics card. This is _not_ what's puzzling me. What is, is
that the spectrum is very often brickwalled at that line: even where the actual valid material is all below 15, 12, 11, 10, 8, or 6 kHz (you'd be surprised how much _does_ have nothing valid above those!), and the
remainder is just uniform noise - it cuts off at the line. Can anyone
think why? It's nowhere near the Nyquist limit of 22050; I could
understand a rolloff _towards_ that to avoid aliasing, and that rolloff
being gentle to avoid other adverse effects, but no, it's brickwalled,
and at the line (which is _well_ below).
J. P. Gilliver <G6JPG@255soft.uk> wrote:
2. The _level_ is often extremely low - especially for some old (say
1960-1999) video clips. (Not all, by any means - but often enough to be
noticeable.) By low, I mean I have to boost them by ×4, or even ×8 or
occasionally ×16, to get the peaks above 50% full scale. (I only use
powers of 2 to avoid distortion.) Is this something YouTube are
imposing? Is audio level adjustment difficult on some common piece of
video capture hardware/software? I even came across one recently where
the uploader _said_ something like "this is quiet, you may have to
adjust" in the notes, so s/he knew about it. This does seem odd.
I suspect people aren't doing a lot of postprocessing: recording from >analogue using their soundcard/etc, uploading the clips to YT, YT does the
transcoding but isn't changing levels. It may be people are using line in >inputs from analogue sources and not setting levels correctly, I don't know.
3. (This is the one that finally prompted me to post.) Far more often
than not - I'd say over 90% of tracks - there's a visible (I can no
longer hear that high) tone around 15½ kHz. I presume in the majority of
cases, it's timebase - 15625 for "PAL" (yes I know, but YKWIM), 15750
for NTSC; even where it's not from an actual video source, I presume it
has picked it up somewhere in the processing, e. g. from a computer
monitor/graphics card. This is _not_ what's puzzling me. What is, is
that the spectrum is very often brickwalled at that line: even where the
actual valid material is all below 15, 12, 11, 10, 8, or 6 kHz (you'd be
surprised how much _does_ have nothing valid above those!), and the
remainder is just uniform noise - it cuts off at the line. Can anyone
think why? It's nowhere near the Nyquist limit of 22050; I could
understand a rolloff _towards_ that to avoid aliasing, and that rolloff
being gentle to avoid other adverse effects, but no, it's brickwalled,
and at the line (which is _well_ below).
Maybe YT have a filter to block timebase frequencies? In the early days >there was a lot of material uploaded from VHS (pre-2010 YT videos are often >240p or similar), which I suspect is where it's coming from on your
examples. I wouldn't be surprised if the timebase leaked onto the audio >track, but contemporary VHS hardware couldn't play it back so people didn't >notice. They might do today, hence a reason to filter it out. And, as
these mega-platforms go, it's easier to have a one-size-fits-all policy than >to do any tailoring to the input material.
Theo
OK, not strictly broadcast, but YouTube is almost a broadcast channel.
I use yt-dlp a lot, usually on default settings (which AIUI usually gets
the best available), and then an extractor set to "extract original audio"
(I use Pazera, as it's easy to be sure it's extracting original without
any further transcoding; however, it's just ffmpeg-based, and I presume
any other similar would yield the same result). [Yes, I looked into using
the audio-only settings for yt-dlp, but they didn't easily lend themselves
to batching; besides, I sometimes _do_ want the video too - clip of an
artist performing, and I want the audio-only one for use in the car. My muscle memory of the keystrokes to extract the audio means I can do it in seconds anyway.] I usually look at the resultant audio - sometimes with
the intention of reducing the filesize, sometimes just out of curiosity.
(I use GoldWave, but I presume almost any other similar utility - such as Audacity - would yield similar results.)
Several observations:
1. The _vast_ majority are coded at 44100 Hz, stereo. I suppose that - "CD quality" - is the default setting for many capture/encoding devices, but
it does seem overkill for mono material, especially of considerable age
(such as from 78s). Still, I'm not surprised. (I very occasionally find
one that _has_ been encoded mono. Though I don't think I've seen any
encoded at less than 44100 - certainly if I have, it's been extremely
rare.)
2. The _level_ is often extremely low - especially for some old (say 1960-1999) video clips. (Not all, by any means - but often enough to be noticeable.) By low, I mean I have to boost them by ×4, or even ×8 or occasionally ×16, to get the peaks above 50% full scale. (I only use
powers of 2 to avoid distortion.) Is this something YouTube are imposing?
Is audio level adjustment difficult on some common piece of video capture hardware/software? I even came across one recently where the uploader
_said_ something like "this is quiet, you may have to adjust" in the
notes, so s/he knew about it. This does seem odd.
3. (This is the one that finally prompted me to post.) Far more often than not - I'd say over 90% of tracks - there's a visible (I can no longer hear that high) tone around 15½ kHz. I presume in the majority of cases, it's timebase - 15625 for "PAL" (yes I know, but YKWIM), 15750 for NTSC; even where it's not from an actual video source, I presume it has picked it up somewhere in the processing, e. g. from a computer monitor/graphics card. This is _not_ what's puzzling me. What is, is that the spectrum is very
often brickwalled at that line: even where the actual valid material is
all below 15, 12, 11, 10, 8, or 6 kHz (you'd be surprised how much _does_ have nothing valid above those!), and the remainder is just uniform
noise - it cuts off at the line. Can anyone think why? It's nowhere near
the Nyquist limit of 22050; I could understand a rolloff _towards_ that to avoid aliasing, and that rolloff being gentle to avoid other adverse
effects, but no, it's brickwalled, and at the line (which is _well_
below).
--
J. P. Gilliver. UMRA: 1960/<1985 MB++G()AL-IS-Ch++(p)Ar@T+H+Sh0!:`)DNAf
I admire him for the constancy of his curiosity, his effortless sense of authority and his ability to deliver good science without gimmicks.
- Michael Palin on Sir David Attenborough, RT 2016/5/7-13
J. P. Gilliver <G6JPG@255soft.uk> wrote:
2. The _level_ is often extremely low - especially for some old (say
1960-1999) video clips. (Not all, by any means - but often enough to be
noticeable.) By low, I mean I have to boost them by ×4, or even ×8 or
occasionally ×16, to get the peaks above 50% full scale. (I only use
powers of 2 to avoid distortion.) Is this something YouTube are
imposing? Is audio level adjustment difficult on some common piece of
video capture hardware/software? I even came across one recently where
the uploader _said_ something like "this is quiet, you may have to
adjust" in the notes, so s/he knew about it. This does seem odd.
I suspect people aren't doing a lot of postprocessing: recording from analogue using their soundcard/etc, uploading the clips to YT, YT does the transcoding but isn't changing levels. It may be people are using line in inputs from analogue sources and not setting levels correctly, I don't
know.
3. (This is the one that finally prompted me to post.) Far more often
than not - I'd say over 90% of tracks - there's a visible (I can no
longer hear that high) tone around 15½ kHz. I presume in the majority of
cases, it's timebase - 15625 for "PAL" (yes I know, but YKWIM), 15750
for NTSC; even where it's not from an actual video source, I presume it
has picked it up somewhere in the processing, e. g. from a computer
monitor/graphics card. This is _not_ what's puzzling me. What is, is
that the spectrum is very often brickwalled at that line: even where the
actual valid material is all below 15, 12, 11, 10, 8, or 6 kHz (you'd be
surprised how much _does_ have nothing valid above those!), and the
remainder is just uniform noise - it cuts off at the line. Can anyone
think why? It's nowhere near the Nyquist limit of 22050; I could
understand a rolloff _towards_ that to avoid aliasing, and that rolloff
being gentle to avoid other adverse effects, but no, it's brickwalled,
and at the line (which is _well_ below).
Maybe YT have a filter to block timebase frequencies? In the early days there was a lot of material uploaded from VHS (pre-2010 YT videos are
often
240p or similar), which I suspect is where it's coming from on your
examples. I wouldn't be surprised if the timebase leaked onto the audio track, but contemporary VHS hardware couldn't play it back so people
didn't
notice. They might do today, hence a reason to filter it out. And, as
these mega-platforms go, it's easier to have a one-size-fits-all policy
than
to do any tailoring to the input material.
Theo
I think you expect too much of people. Most of those with footage tend to do >what they always do and just put that up. If you really wanted to do the >processing its easy enough, but most just don't bother. I tend to know if I
get anything off line be it podcast, Youtube or whatever and use Goldwave
to fix stuff and resave it at the bit rate I like. If you want to hear brick
wall filtering, any pop concert recently recorded by the bbc is like that. >After all FM never had anything above 15khz except noise most of the time.
I have a few custom effects saved in Goldwave, one is superfast gain
update, which effectively compresses the dynamic range. There are a few
that only compress peaks, and some for matching the upper levels without
clipping. I also have quite a few parametric settings like presence reduce, >and tizzy enhance for those under the blanket recordings. There are very >light touch noise reductions from cclipboard as well, and some custom >pop/click ones to clean up crackles.
You'd think they'd set the brick wall to remove timebase whistle, if
that's the case, though.
On 05/08/2023 12:01, J. P. Gilliver wrote:
You'd think they'd set the brick wall to remove timebase whistle, if
that's the case, though.
What frequency of line whistle, though? On the current video streaming >services, line whistle may be as low as 10 kHz or as high as 20 kHz, >depending on the original video standard. While the high end isn't
likely to be a problem, the lower end may mask wanted signals.
It's not a new problem, one very famous album, much admired for its
audio quality, has a constant scan whistle on most of the tracks from
the monitors used on the studio computers.
3. (This is the one that finally prompted me to post.) Far more often
than not - I'd say over 90% of tracks - there's a visible (I can no
longer hear that high) tone around 15½ kHz. I presume in the majority of cases, it's timebase - 15625 for "PAL" (yes I know, but YKWIM), 15750
for NTSC; even where it's not from an actual video source, I presume it
has picked it up somewhere in the processing, e. g. from a computer monitor/graphics card. This is _not_ what's puzzling me. What is, is
that the spectrum is very often brickwalled at that line: even where the actual valid material is all below 15, 12, 11, 10, 8, or 6 kHz (you'd be surprised how much _does_ have nothing valid above those!), and the remainder is just uniform noise - it cuts off at the line. Can anyone
think why? It's nowhere near the Nyquist limit of 22050; I could
understand a rolloff _towards_ that to avoid aliasing, and that rolloff being gentle to avoid other adverse effects, but no, it's brickwalled,
and at the line (which is _well_ below).
On Friday, 4 August 2023 at 02:29:02 UTC+1, J. P. Gilliver wrote:
3. (This is the one that finally prompted me to post.) Far more often
than not - I'd say over 90% of tracks - there's a visible (I can no
longer hear that high) tone around 15½ kHz. I presume in the majority of
cases, it's timebase - 15625 for "PAL" (yes I know, but YKWIM), 15750
for NTSC; even where it's not from an actual video source, I presume it
has picked it up somewhere in the processing, e. g. from a computer
monitor/graphics card. This is _not_ what's puzzling me. What is, is
that the spectrum is very often brickwalled at that line: even where the
actual valid material is all below 15, 12, 11, 10, 8, or 6 kHz (you'd be
surprised how much _does_ have nothing valid above those!), and the
remainder is just uniform noise - it cuts off at the line. Can anyone
think why? It's nowhere near the Nyquist limit of 22050; I could
understand a rolloff _towards_ that to avoid aliasing, and that rolloff
being gentle to avoid other adverse effects, but no, it's brickwalled,
and at the line (which is _well_ below).
AFAIU the MP4 (-f 140 is generally the highest quality in that
codec) audio downloads have a frequency roll-off at around that
frequency (16kHz iirc). This can be seen in audacity using the
spectrogram setting; right-click on the vertical kHz bar and
select "zoom to fit"). I have started to download audio in OPUS
(-f 251) as it doesn't seem to have this roll-off. Then
batch-converting files with FFMPEG (thank you to those in here
who helped me with this in the past!) to MP3 for wider
compatibility with my various media players.
Sysop: | Keyop |
---|---|
Location: | Huddersfield, West Yorkshire, UK |
Users: | 307 |
Nodes: | 16 (2 / 14) |
Uptime: | 28:28:54 |
Calls: | 6,907 |
Calls today: | 1 |
Files: | 12,376 |
Messages: | 5,427,675 |
Posted today: | 1 |