Digital recording 101 wanted

[Marc Heusser]

I've finally switched from mini-disc to directly encoding mp3 and am
very happy with it :-)
Now I do need a digital recording 101, since I can now switch off
automatic level control with my Marantz PMD-670 , and I'd like to do so.
I'm recording from a condensor microphone.

What I want to do:
- record voices without automatic volume control, since I'd like to
preserve the differences in volume of the different voices.

Am I correct in assuming that 0 dB should never be reached, because it
will be clipped?
When transferring recordings from mini-disc (I know, never encode twice,
but I'll get rid of mini-disc), I had recordings with clipped sounds,
and it was bad :-(.
(clipped meaning limited by being at the maximum digital value)

Or do you even keep maximum sound level (not averaged but single value)
below 0 dB, and if so, how much?
(assuming 0 dB on the display of the PMD-670 is the maximum value the
recorder can encode - the level meters are marked from up to 0 dB and a
separate indicator for over.)

Thanks in advance

Marc

(PS I'm an electronics engineer, so feel free to throw technical
explanations at me)

--
Marc Heusser - Zurich, Switzerland
Coaching - Consulting - Counselling - Psychotherapy
http://www.heusser.com
remove the obvious CHEERS and MERCIAL... from the reply address
to reply via e-mail

[Abyssmal]

On Mon, 06 Oct 2003 16:07:05 +0200, Marc Heusser
lid> wrote:


>Am I correct in assuming that 0 dB should never be reached, because it
>will be clipped?

Yes.


>Or do you even keep maximum sound level (not averaged but single value)
>below 0 dB, and if so, how much?

I usually aim for -3 when recording, to give me a little room to add
processing later.

If a vocalist has good mic technique, I can bring him in at that level
with little fear of going over. For less than stellar vocalists, I may
compress/limit them to avoid clipping.

Randall

[David Satz]

Marc Heusser wrote:

> Am I correct in assuming that 0 dB should never be reached, because it
> will be clipped?

The "Over" light should never light, because that definitely indicates
clipping.

For most other types of indicator, some integration time greater than one
sampling interval is involved, so single-sample peaks don't register fully.
They can be reading 4 dB below full scale while the actual single-sample
value is right at the limit. This is particularly true in live recording
with high-quality microphones placed reasonably close to the sound source.

(I've directly observed this; Sony TCD-D7 DAT recorders can read 5 dB lower
than single-sample peak values and the TCD-D10 Pro can read 4 dB below.)

There are digital meters which respond to single-sample peaks accurately,
but they tend to cost hundreds of dollars just by themselves. Even the
"overload" indicators on most digital recording equipment will not flash
unless three successive samples are in overload--so I really mean that
the "Over" indicator should never come on in any serious recording.


> Or do you even keep maximum sound level (not averaged but single value)
> below 0 dB, and if so, how much?

Unless you record in eerily quiet surroundings, you rarely need a full
16-bit dynamic range. So it won't hurt if you leave, say, 3 - 5 dB of
unused range at the very top. This gives you some security in case the
live recording gets a little louder than you had expected, and it also
leaves some slack for subsequent signal processing which will sometimes
have different (or more complex) headroom limits.

One concern that I have never quite resolved is the fact that there is a
worst-case input (sequence of sample values) for any given set of digital
filter coefficients, such that all the bins will be contributing their full-
scale output to each output value. If a signal resembling that worst-case
pattern should happen to occur in a recording, the filter will most likely
"clip" digitally or even wrap around (double-yuck).

If the filter designer scales the entire design downwards to avoid this
problem, he/she surrenders 6 - 8 dB of s/n advantage that would usually
be safe, apart from a horrible coincidence or a highly determined bench
test. But not scaling the signal (or the filter coefficients) downward
means leaving a distinct possibility of severe clipping and/or wraparound.

Since this issue is handled individually in each filter design, I like to
allow at least 1 - 2 dB of unused dynamic range when I make a recording
that's intended for compact disc, since I never know what the parameters
of the filters will be on the playback side. 3 dB would be even better,
but by that point I start to worry that the CD will be perceived as being
too "soft" compared to other people's--you don't want to make people have
to fiddle with their volume control, heaven forbid (we all know that
volume, balance and tone controls were made to be looked at, not used).

[David Satz]

Marc Heusser wrote:

> Am I correct in assuming that 0 dB should never be reached, because it
> will be clipped?

The "Over" light should never light, because that definitely indicates
clipping.

For most other types of indicator, some integration time greater than one
sampling interval is involved, so single-sample peaks don't register fully.
They can be reading 4 dB below full scale while the actual single-sample
value is right at the limit. This is particularly true in live recording
with high-quality microphones placed reasonably close to the sound source.

(I've directly observed this; Sony TCD-D7 DAT recorders can read 5 dB lower
than single-sample peak values and the TCD-D10 Pro can read 4 dB below.)

There are digital meters which respond to single-sample peaks accurately,
but they tend to cost hundreds of dollars just by themselves. Even the
"overload" indicators on most digital recording equipment will not flash
unless three successive samples are in overload--so I really mean that
the "Over" indicator should never come on in any serious recording.


> Or do you even keep maximum sound level (not averaged but single value)
> below 0 dB, and if so, how much?

Unless you record in eerily quiet surroundings, you rarely need a full
16-bit dynamic range. So it won't hurt if you leave, say, 3 - 5 dB of
unused range at the very top. This gives you some security in case the
live recording gets a little louder than you had expected, and it also
leaves some slack for subsequent signal processing which will sometimes
have different (or more complex) headroom limits.

One concern that I have never quite resolved is the fact that there is a
worst-case input (sequence of sample values) for any given set of digital
filter coefficients, such that all the bins will be contributing their full-
scale output to each output value. If a signal resembling that worst-case
pattern should happen to occur in a recording, the filter will most likely
"clip" digitally or even wrap around (double-yuck).

If the filter designer scales the entire design downwards to avoid this
problem, he/she surrenders 6 - 8 dB of s/n advantage that would usually
be safe, apart from a horrible coincidence or a highly determined bench
test. But not scaling the signal (or the filter coefficients) downward
means leaving a distinct possibility of severe clipping and/or wraparound.

Since this issue is handled individually in each filter design, I like to
allow at least 1 - 2 dB of unused dynamic range when I make a recording
that's intended for compact disc, since I never know what the parameters
of the filters will be on the playback side. 3 dB would be even better,
but by that point I start to worry that the CD will be perceived as being
too "soft" compared to other people's--you don't want to make people have
to fiddle with their volume control, heaven forbid (we all know that
volume, balance and tone controls were made to be looked at, not used).

[Arny Krueger]

"Marc Heusser" lid>
wrote in message

> I've finally switched from mini-disc to directly encoding mp3 and am
> very happy with it :-)
> Now I do need a digital recording 101, since I can now switch off
> automatic level control with my Marantz PMD-670 , and I'd like to do
> so. I'm recording from a condensor microphone.
>
> What I want to do:
> - record voices without automatic volume control, since I'd like to
> preserve the differences in volume of the different voices.
>
> Am I correct in assuming that 0 dB should never be reached, because it
> will be clipped?

Right.

> When transferring recordings from mini-disc (I know, never encode
> twice, but I'll get rid of mini-disc), I had recordings with clipped
> sounds, and it was bad :-(.

Yup, clipping is bad, and to be avoided. It's far better to get only
middling peak levels than to have clipping.

> (clipped meaning limited by being at the maximum digital value)

> Or do you even keep maximum sound level (not averaged but single
> value) below 0 dB, and if so, how much?

No, keep *all* peaks well under 0 dB.

[Arny Krueger]

"Marc Heusser" lid>
wrote in message

> I've finally switched from mini-disc to directly encoding mp3 and am
> very happy with it :-)
> Now I do need a digital recording 101, since I can now switch off
> automatic level control with my Marantz PMD-670 , and I'd like to do
> so. I'm recording from a condensor microphone.
>
> What I want to do:
> - record voices without automatic volume control, since I'd like to
> preserve the differences in volume of the different voices.
>
> Am I correct in assuming that 0 dB should never be reached, because it
> will be clipped?

Right.

> When transferring recordings from mini-disc (I know, never encode
> twice, but I'll get rid of mini-disc), I had recordings with clipped
> sounds, and it was bad :-(.

Yup, clipping is bad, and to be avoided. It's far better to get only
middling peak levels than to have clipping.

> (clipped meaning limited by being at the maximum digital value)

> Or do you even keep maximum sound level (not averaged but single
> value) below 0 dB, and if so, how much?

No, keep *all* peaks well under 0 dB.

[Marc Heusser]

In article >,
(David Satz) wrote:

> The "Over" light should never light, because that definitely indicates
> clipping.
>
> For most other types of indicator, some integration time greater than one
> sampling interval is involved, so single-sample peaks don't register fully.
> They can be reading 4 dB below full scale while the actual single-sample
> value is right at the limit. This is particularly true in live recording
> with high-quality microphones placed reasonably close to the sound source.
>
> (I've directly observed this; Sony TCD-D7 DAT recorders can read 5 dB lower
> than single-sample peak values and the TCD-D10 Pro can read 4 dB below.)
>
> There are digital meters which respond to single-sample peaks accurately,
> but they tend to cost hundreds of dollars just by themselves. Even the
> "overload" indicators on most digital recording equipment will not flash
> unless three successive samples are in overload--so I really mean that
> the "Over" indicator should never come on in any serious recording.

Thank you all for the explanations and experiences, that saves a lot of
testing.

As for the level meters, I'll test.
I have some hopes as the meters of the PMD-670 do indicate both actual
level (averaged I guess) and peak, and in addition it has a digital
indicator for maximum peak level (eg -2dB) up to some point in time that
will only be reset by pushing a button (kind of nice, isn't it). I'd
hope it will store even single-sample value peaks.

Marc

--
Marc Heusser - Zurich, Switzerland
Coaching - Consulting - Counselling - Psychotherapy
http://www.heusser.com
remove the obvious CHEERS and MERCIAL... from the reply address
to reply via e-mail

[Marc Heusser]

In article >,
(David Satz) wrote:

> The "Over" light should never light, because that definitely indicates
> clipping.
>
> For most other types of indicator, some integration time greater than one
> sampling interval is involved, so single-sample peaks don't register fully.
> They can be reading 4 dB below full scale while the actual single-sample
> value is right at the limit. This is particularly true in live recording
> with high-quality microphones placed reasonably close to the sound source.
>
> (I've directly observed this; Sony TCD-D7 DAT recorders can read 5 dB lower
> than single-sample peak values and the TCD-D10 Pro can read 4 dB below.)
>
> There are digital meters which respond to single-sample peaks accurately,
> but they tend to cost hundreds of dollars just by themselves. Even the
> "overload" indicators on most digital recording equipment will not flash
> unless three successive samples are in overload--so I really mean that
> the "Over" indicator should never come on in any serious recording.

Thank you all for the explanations and experiences, that saves a lot of
testing.

As for the level meters, I'll test.
I have some hopes as the meters of the PMD-670 do indicate both actual
level (averaged I guess) and peak, and in addition it has a digital
indicator for maximum peak level (eg -2dB) up to some point in time that
will only be reset by pushing a button (kind of nice, isn't it). I'd
hope it will store even single-sample value peaks.

Marc

--
Marc Heusser - Zurich, Switzerland
Coaching - Consulting - Counselling - Psychotherapy
http://www.heusser.com
remove the obvious CHEERS and MERCIAL... from the reply address
to reply via e-mail