Topic: How to determine proper volume levels when sampling and mastering???

[mysterymadman]

Hi,

I'm pretty much a newbie to taping and mastering, so I hope people will bear with me on this one. Also, it would not surprise me if the questions I pose down below will already have been asked (partially or in full) here, and/or if they are very basic, so I hope I'm not flogging a dead horse, and if I am, I hope someone can point me to (a) similar thread(s), or a tutorial even....

Also, I'm sorry that this item has become rather long. The questions themselves are probably quick to answer though, so I hope you guys can help me with them (even if your answers are much shorter than my questions are). 

Well then, the questions (which are posed towards the end of this item) I have are undoubtedly straightforward for you guys, but I am just unsure as to what gives best results, and your opinions and experience is of great value to me to get the best possible results. In brief, they involve properly setting and managing volume levels when sampling and mastering recordings.

Firstly, the situation:
I (stealthily) taped the Iron Maiden concert in The Netherlands on November 27th 2006, and despite having done my best to tape it from a sweet spot, from my position it was very "bass-heavy", and the audience peaks are very loud. During practice rehearsals of our daughter's band at the local school of music, I found out that at that particular practice room the drums (esp. the snare) tend to be twice as loud as the rest of the instruments. I then experimented with the sampling volume, and basically used two main approaches:
1) Make sure no clipping occurred: this kept the drum peaks within bounds, but seemed to muffle the rest of the instruments and music.
2) Sample at a higher volume, ignoring the fact that the drums were clipped: this seemed to give an overall more pleasing result, and the clipping was not audible.
I considered doing the same for the IM recording, but I'm not certain matters are quite the same this time. I again made two raw samples, one without clipping, and one with clipping (of the audience burst, and possibly/probably also of some of the louder bass parts). This time though, I *think* the low volume one ought to be better, as the inbalance between the low frequency peaks and the rest is of course not as much off as with the practice rehearsal recordings.
Then, based on what I did for the practice rehearsals, for the IM recordings I initially used the louder (i.e. clipped) recordings, and used two passes of equalisation filters to get a more pleasing end mix. However... when taking a better look at this, during each step more and more clipping must have occurred. Firstly, the raw sample was already clipped, then the first eq. pass reduced the low freqs and boosted the higher ones, finally an additional eq. pass boosted the mid and higher ones quite somewhat more.
Of course not having used a negative output gain during the eq. passes must have been a killer, and will undoubtedly have resulted in a 'less than optimal' sound quality due to rather extreme clipping that is stacked on top of each other at each eq. pass...

So then, on to the questions:
1) Is it -theoretically and/or practically spoken- best to always make sure no clipping occurs when sampling (as I would expect), even if one frequency range is significantly louder than the rest, or can clipping be acceptable?
(note: I fear this may not even be such a straightforward question, as my gut feeling suggests to me that it depends on the situation and the inbalance. For the drum peaks it worked fine to allow some clipping, but these are very short and sharp peaks, whereas the 'bass peaks' surely are not. I'd guess in this case clipping is a bigger issue, right?).
2) Overall spoken (depending on the answer to question nr. 1, I guess), what would be an 'ideal' sample volume?
(my guess: choose as loud a volume as possible, without clipping, and keeping the loudest peaks below say -1 dB or so, right?)
3) I seem to recall having read that ideally average volume levels should be around -12 dB. Correct?
4) To compress or not? If I trim the output gain to a level such that the average volume level 'dances' around -12dB, with peaks going up to levels as loud as -3dB (or  more even), will that result in an undesirable sound, or is it o.k. if the volume levels fluctuate that heavily between the averages and the peaks?
(I read http://taperssection.com/index.php?topic=56640 and guess that no compression should be necessary. Right?)
IOW: should that be compressed then, or does that result in loss of dynamics or so???
5) I would guess that the situation as described under question 4 is undesirable, and that indeed it should be 'soft limited' to -say- -12dB then. Is that indeed correct, or are there other things you guys would do, or perhaps nothing at all even?
6) I use Sony's Sound Forge v8. Now, when using the 'Sony graphic equaliser', and boosting specific ranges to -say- +10dB, and then reducing the output gain to -say- -10dB or so too, will that properly avoid the boosted peaks being clipped, or should I first reduce the output gain by setting it to -10dB, and then apply the filter?
(Note: judging from the volume bars in the preview, I'd guess this combination of the two works fine, and will not introduce any clipping, but I'm asking just to be sure....).
7) Anything else I'm missing?

Thanks a lot in advance for taking the time to read (and hopefully answer) these questions, and cheers!
MM

[mysterymadman]

Hi again,

As I did not receive any replies to my initial questions, I decided to also ask a former colleague of mine, who is very active in stuff like this. He's a guitar teacher and has spent a great amount of hours on sampling and mastering too. He did mention a base strategy in a few lines, and I hope those of you who are also uncertain about this can benefit from it, and/or that those of you who do know can perhaps comment on it to let me (and other readers) know if indeed the below (esp. the question at the end) is correct...

Basically what he wrote is:
1) Sample without clipping, keeping the peaks just below 0dB.
2) Normalise up to 0dB (NOT up to -12dB, as nowadays seems to be old-fashioned -or so he said-).
3) Equalise (and/or make other edits), adjusting the output gain such that the 0dB border will not be crossed.
4) Normalise up to 0dB again.
5) Cut the recordings.

Now, please note that he did not explicitly mention two normalise passes, but just one. Yet, that is something I added myself, such that both recordings (in my case) are normalised against the same peak sound level.

Last night I did all that, and that way I got a sound which I liked a lot (using the 'DAT direct' on my stereo, looping directly through to it in Soundforge's preview mode of the equaliser, and then slightly adjusting the settings of it such that it sounds good to me).

Now........I did theorise about one further possible optimisation, which I hope you guys can comment on: in the raw recordings, some 'gunshot-like' (or 'clap-like', if you will) clipping occurs. I could of course first remove (or lower the volume of) ALL of those peaks, and then perform the normalisation. I would think that should result in a better dynamic range, but maybe this is hardly audible at all, whilst requiring a lot of effort....

What say you?

Cheers and tnx in advance!
MM

[yousef]

Hello again...

I'd say - don't do the 'first-pass' normalization, it just increases the risk of clipping when you do any subsquent eq'ing. I would do all editing and eq first and save normalization until your last step.

As for peaks caused by clapping/audience noise - this has been discussed very recently on TS (within the last couple of days) but I can't seem to find where. For what it's worth, I've had very good results using Audacity's volume envelope function to gently fade audience levels between songs to an acceptable level. You would then normalize to 0dB or whatever your preference is.

This means that you not only avoid the annoying clipping caused by clapping but if you are recording a quieter source, you aren't normailizing to a level dictated by how loud the audience is.

Audacity is free to download and the volume envelope thing is very intuitive and flexible.

Others would discard all this and suggest judicious use of compression but that is well beyond my ken. Of course, I may have misunderstood you - if your troublesome peaks are caused by Iron Maiden rather than the audience, compression is the way to go. This might be a good place to start:

http://taperssection.com/index.php?topic=68748.0

Hope this helps.

Yousef

[mysterymadman]

As for peaks caused by clapping/audience noise - this has been discussed very recently on TS (within the last couple of days) but I can't seem to find where. For what it's worth, I've had very good results using Audacity's volume envelope function to gently fade audience levels between songs to an acceptable level. You would then normalize to 0dB or whatever your preference is.

This means that you not only avoid the annoying clipping caused by clapping but if you are recording a quieter source, you aren't normailizing to a level dictated by how loud the audience is.

Others would discard all this and suggest judicious use of compression but that is well beyond my ken.

Hi Yousef,

Tnx for helping me out once more with my questions, that's very much appreciated!

So basically it is as I figured: having the clippings dictate the normalisation, does indeed have an influence...
Now, the tricky thing is that whereas most of the times the volume between songs could easily be lowered, there is also a HUGE audience outburst during the beginning of some 4 songs or so... I could of course somehow either compress that, or just plainly lower the volume of that, but the thing which may then happen, is that perhaps it sounds weird, as not all of the recording of the music itself is then treated the same way. But then again, perhaps that is exactly what is needed...

In itself, I had decided (trying) to avoid compressing the sound, as that by definition would imply less dynamics... But then (again), perhaps slightly soft limiting the audience outbursts alone is not such a bad idea after all... 

Alrighty, I'll have to have a go at this, but before I do I think I'll first burn the current versions to CDs and will try them on several different stereos to see just how it sounds on 'real-life' systems. If indeed I consider the recordings to lack dynamics, or be otherwise unpleasing, I'll try some of those strategies...

At least, for now I know my assumptions were correct, and that acting accordingly could possibly indeed improve matters... 

Tnx and cheers!
MM

[yousef]

Now, the tricky thing is that whereas most of the times the volume between songs could easily be lowered, there is also a HUGE audience outburst during the beginning of some 4 songs or so... I could of course somehow either compress that, or just plainly lower the volume of that, but the thing which may then happen, is that perhaps it sounds weird, as not all of the recording of the music itself is then treated the same way. But then again, perhaps that is exactly what is needed...

Audacity allows you to do smooth fades so that (although there will obviously be an audible change in level) the effect will be a little more gradual and natural than a sudden change. It's not the ideal situation but I think the art of compromise is a large part of successful stealth recordings - if you aren't the sort of person who wants to leave recordings entirely "as-is", you have to decide what solution sounds most acceptable to your ears. Most times, I'd go for the fade-in but I imagine that it wouldn't suit all types of music.

Good luck!

[Church-Audio]

As for peaks caused by clapping/audience noise - this has been discussed very recently on TS (within the last couple of days) but I can't seem to find where. For what it's worth, I've had very good results using Audacity's volume envelope function to gently fade audience levels between songs to an acceptable level. You would then normalize to 0dB or whatever your preference is.

This means that you not only avoid the annoying clipping caused by clapping but if you are recording a quieter source, you aren't normalizing to a level dictated by how loud the audience is.

Others would discard all this and suggest judicious use of compression but that is well beyond my ken.

Hi Yousef,

Tnx for helping me out once more with my questions, that's very much appreciated!

So basically it is as I figured: having the clippings dictate the normalisation, does indeed have an influence...
Now, the tricky thing is that whereas most of the times the volume between songs could easily be lowered, there is also a HUGE audience outburst during the beginning of some 4 songs or so... I could of course somehow either compress that, or just plainly lower the volume of that, but the thing which may then happen, is that perhaps it sounds weird, as not all of the recording of the music itself is then treated the same way. But then again, perhaps that is exactly what is needed...

In itself, I had decided (trying) to avoid compressing the sound, as that by definition would imply less dynamics... But then (again), perhaps slightly soft limiting the audience outbursts alone is not such a bad idea after all... 

Alrighty, I'll have to have a go at this, but before I do I think I'll first burn the current versions to CDs and will try them on several different stereos to see just how it sounds on 'real-life' systems. If indeed I consider the recordings to lack dynamics, or be otherwise unpleasing, I'll try some of those strategies...

At least, for now I know my assumptions were correct, and that acting accordingly could possibly indeed improve matters... 

Tnx and cheers!
MM

One thing I do is, Separate all the songs into individual tracks. Then I do my eq and or compression first. Then I normalize all the files again to smooth out the transitions from song to song. This gives me a good average level for all the tracks, you can not normalize a whole show on one pass. You have to take the recording apart and do it song by song. This is a mistake many people make because lets say you have one song where there is a loud spike but in the next song there is not the Normalization will bring everything down because of the one spike. It averages the peeks and uses RMS in some cases to do the math on how loud it can boost.

So doing separate songs gives you hotter levels closer to the golden 0db. And always allow the program to scan the wav file first, so it can get the average levels. Never blindly say Normalize to 0db, if you do that your bound to get distortion and a less then stellar recording.

Chris Church

[Brian Skalinder]

One thing I do is, Separate all the songs into individual tracks. Then I do my eq and or compression first. Then I normalize all the files again to smooth out the transitions from song to song. This gives me a good average level for all the tracks, you can not normalize a whole show on one pass. You have to take the recording apart and do it song by song. This is a mistake many people make because lets say you have one song where there is a loud spike but in the next song there is not the Normalization will bring everything down because of the one spike. It averages the peeks and uses RMS in some cases to do the math on how loud it can boost.

The issue you'll run into normalizing each individual file:  you lose the relative dynamic range across the performance.  So a particularly quiet song will sound just as loud as the loudest song.  If there's an outlying peak, I think the best way to handle it is to first compress just the outlying peak, then apply compression across the complete recording.

1) Is it -theoretically and/or practically spoken- best to always make sure no clipping occurs when sampling (as I would expect), even if one frequency range is significantly louder than the rest, or can clipping be acceptable?

Don't clip.  It can, though doesn't always, sound ugly.  And editing clipped audio in post will often present artifacts, even if the clipping isn't audible in the untouched master.
 

2) Overall spoken (depending on the answer to question nr. 1, I guess), what would be an 'ideal' sample volume?
(my guess: choose as loud a volume as possible, without clipping, and keeping the loudest peaks below say -1 dB or so, right?)

Yup, you've got it - peak as close to 0 dB as possible, without hitting 0 dB.

4) To compress or not? If I trim the output gain to a level such that the average volume level 'dances' around -12dB, with peaks going up to levels as loud as -3dB (or  more even), will that result in an undesirable sound, or is it o.k. if the volume levels fluctuate that heavily between the averages and the peaks?
(I read http://taperssection.com/index.php?topic=56640 and guess that no compression should be necessary. Right?)
IOW: should that be compressed then, or does that result in loss of dynamics or so???

The difference between average and peaks will depend on the dynamic range of the performance.  Don't sweat it.  It is what it is.  Just set your peaks properly.  If you want to compress to reduce dynamic range later, you can, but that's not something to worry about , the averages will The point of compression is to reduce dynamic range.  Really depends on your personal preference, playback system, etc.

5) I would guess that the situation as described under question 4 is undesirable, and that indeed it should be 'soft limited' to -say- -12dB then. Is that indeed correct, or are there other things you guys would do, or perhaps nothing at all even?

Again, just set your peaks properly.  If, upon playback, you find the dynamic range too broad, you can compress in post for future listening.

6) I use Sony's Sound Forge v8. Now, when using the 'Sony graphic equaliser', and boosting specific ranges to -say- +10dB, and then reducing the output gain to -say- -10dB or so too, will that properly avoid the boosted peaks being clipped, or should I first reduce the output gain by setting it to -10dB, and then apply the filter?

I'd be surprised if you needed to increase any part of the frequency range by +10 dB.  I vaguely recall someone saying here it's better to take away than to add gain with EQ.  I do very little EQ, though, so...can't add much more to this one.

[mysterymadman]

Hi guys

Tnx a lot for the answers!
It has taken a while for me to get around to answering, and editing audio again, as I've been extremely busy with moving to a newly built house, and in between moving, painting, etc., I have hardly used the computer in my spare time.
Anyway... most of that is over with now, and over the past two weeks I have finally started to have some time again to get back to this.

Sooo, your advice is clear, and after I had posted the initial questions I myself experimented around a bit too. The strategy I came up with is:
1) Sample as close as possible to 0dB, without introducing clipping by the sampling process.
2) Remove any clipping and/or other loud/undesired peaks that may be present.
3) If necessary, use an equaliser pass, if possible with as subtle boosts/reductions as possible, and using output gain on the entire recording in order to compensate for the max. boost (or reduction, if no boost is used).
4) Normalise to 0dB.
5) Perform other edits if so desired (like shortening unnecessarily long audience parts, e.g.).
6) Cut.

I have tried this on a few recordings now, and this works pretty well, I think.

Then, regarding the occasional loud peaks that are not clipped: I found an *excellent* filter for that, or more correctly, two actually.
1) Dynamics: this can be very nicely configured to reduce loud plosives etc.
2) Wave Hammer: a very nice one! Allows you to set thresholds for a compressor and/or volume maximiser.

Both of these effects are highly configurable, and in combination with the real-time preview mode (of unlimited length!) is a very powerful way to clear up rough parts of the recordings, as well as some simply loud peaks.

At present, I'm remastering my Depeche Mode recordings, which I had previously mastering in an incorrect way (first ever recording, so the inexperience cut in...). It is almost done now and it sounds very pleasing to me. The first 3 songs were played so loudly over the PA though, that several parts of them are blown to smithereens on my recording, and sound almost like they're played over a circular saw or so. (when looking at the waveforms, they shoot all over the dB scale)
These horrendous parts can be very much improved with both the above mentioned filters, and I will try to find optimal settings for them this week. The volume of most of the recording now lies within the -6dB band, with only very little shooting over it. Hence, I'm considering using a wave hammer on, which will reduce the maximum peaks to -6dB (once all individual parts are done to my satisfaction), and then normalising the lot to 0dB and cutting it.

Cheers,
MM