• Register
  • Help
Page 1 of 2 12 LastLast
Results 1 to 10 of 13

Topic: Creating samples: To normalize or not to normalize.

  1. #1

    Creating samples: To normalize or not to normalize.

    I\'m making my own multi-velocity .gig files, should I normalize all of the wav files before I build the instrument, or should the softer velocities be quieter in the .wav file?

  2. #2

    Re: Creating samples: To normalize or not to normalize.

    It depends on how quiet the samples are. If they are really low, then you are losing your s/n ratio (and might as well record in 8bits). However, if your leves are reasonably good, I would go with the natural amplitude of the instrument. It also depends on the instrument itself....can you give me a bit more details?

  3. #3

    Re: Creating samples: To normalize or not to normalize.

    From a sonic purity standpoint, you want all samples to be at the maximum level (number of bits) possible. BUT, if you have to move the mic or change gain to accomodate this, the samples won\'t match timbrally...

    ALSO - normalizing also destroys sonic purity. Personally, I try to never use normalization except as a grunge kind of effect. Record your samples as hot as possible.

    I don\'t really have a rational answer, I\'m just throwing more mud on the question... [img]images/icons/rolleyes.gif[/img]


  4. #4

    Re: Creating samples: To normalize or not to normalize.

    \"ALSO - normalizing also destroys sonic purity.\"
    AAAAh! Wrong! Normalizing does not change the sound quality at all (it is NOT compression). It only increases the overall volume by the amount from the loudest/highest wave peak sample to the point short of clipping. Like a magnifing glass, the wave only gets bigger, but does not change shape. If you try to normalize a wave that already clips or reaches the point just short of clipping, nothing will happen and nothing changes. Compression is a different matter.

    And oh yes. Gosh, too much overhead. What a terrible problem! [img]images/icons/rolleyes.gif[/img]

    Also, if you DO normalize, remember that noise also gains in volume. Use something like Cool Edit to get a noise signature and remove the noise, before or after normalizing.

  5. #5

    Re: Creating samples: To normalize or not to normalize.

    OK, I\'ll jump in with an opinion on this one. I\'ve tried to hear a problem after a file has been normalized and just can\'t hear one. So I\'ve adopted the practice of normalizing all samples and restoring the note-to-note balance in the editor. You generally have to tweak these anyway so it\'s not extra work. I find that my region attenuation settings take on an overall pattern from instrument to instrument and the master attenuation in the instrument properties becomes a better overall indicator of what the relative volume will be.

    All that said, I agree that samples should be recorded as hot as practical and processed (normalized, EQ\'d, etc.) as little as possible because each pass of processing, while it may correct a specific deficiency, raises the risk of producing a sonic cadaver.

  6. #6

    Re: Creating samples: To normalize or not to normalize.

    Wow, thanks for all the input. All I\'m doing is making some ensemble patches out of various and sundry solo instruments. I\'ve layered things up to make a horn section from solo instruments and it takes up about 40 voices, so I\'m merely re-recording the wavs down so that it only takes two per note. I went ahead and normalized them as best as I could, becuase of the way I made my \"ensembles\" some notes, like the high notes, were considerably louder than the rest, so I had to do some kind of balancing.

    Having normalized all the wavs I then, quite obviously, didn\'t get much dynamic change over the velocities, so I used the editor to put in attenuation on those layers, and it worked out well. Except for needing to normalize to balance the wavs within one velocity, I guess my 2nd step deafeated the 1st. ;-)

  7. #7

    Re: Creating samples: To normalize or not to normalize.

    Depends on the instrument and recording process.

    It is my firm opinion that some sort of normalizing should be done. One for overall gain of the loudest sample. And one for consistancy between velocities.

    Anyhow, this doesn\'t mean normalize every sample to be -1db. Waht you should probably do is a \"batch\" normalize, or in wavelab its called Meta normalize. This will scan ALL files in the batch and then adjsut all samples to make the LOUDEST sample at the desired level, and keep the natural dynamic differences between all teh samples.

    Another thing one can do (I like but other people dont) is RMS, or percieved loudness normalizing. Do this over a group of each velocitiy level. You\'ll get a more consistant level in the playback. This however isn\'t \"natural\", so purists may disagree. Its all up to the way you record tho. Not every player will be perfectly the same every note.....is it the instrument...or is it the player....or is it the room?

  8. #8

    Re: Creating samples: To normalize or not to normalize.

    I was once told not to normalise sounds because it \'screwed up\' sounds. Prior to that I\'d always thought that all normalising did was take the loudest peak in a sound and change the gain of the whole sound so that the peak was at, say, 0db.

    So I did a quick and dirty check by putting some music into wavelab and distributing twenty random markers through the file. Then I added a marker at the peak point. I then wrote down the signal level at every marker. Then I normalised the file. The peak went from -8db to 0db. The levels at the markers all went up by exactly the same amount. So the dynamics within the signal hadn\'t been changed.

    The way I see it, you can use normalising and it won\'t screw up the signal of the individual wav file being processed.

    However, there are three other factors to take into account:

    1. If you don\'t use \'meta\' type normalising you will lose the relative level differences between your samples. Theoretically you could and up with each wav in your multi velocity split at the same volume. If you use normalising and want to be sure your dynamics are correct, note the levels of each sample before you normalise them, and then use Gigastudio\'s attenutation to reintroduce the relative volume differences once you\'ve assigned the samples.

    2. Depending on the sound, a virtually inaudible click can reduce the effectiveness of normalising. Sometimes you\'ll see a waveform which seems pathetically low level, but which doesn\'t seem to \'gain\' anything with normalising. Chances are that there is a spurious transient in the signal which is sitting near 0db in level, preventing the file from benefitting much from normalising.

    3. Donnie pointed something out a while ago that I hadn\'t thought through. Some sounds seem soft when you look at your metering, but in fact cut through very well. This is sometimes because the sound has a lot of content at a frequency to which human ears are very sensitive. Because meters aren\'t usually designed to react like ears, they don\'t show this \'perceived\' loudness. Donnie is loathe to do much normalising because he\'d prefer to maintain any dynamics which are a result of timbral variance, as opposed to peaks in signal \'level\'.

    In the end, it\'s horses for courses. Some sounds don\'t want to be touched, some handle normalising really well, and others even need to be seriously compressed before they resemble anything which people will recognise.

    If it sounds good - it IS good [img]images/icons/grin.gif[/img]

  9. #9

    Re: Creating samples: To normalize or not to normalize.

    If you have recorded 24 bit and need to convert to 16 bit, it is a good idea to normalize before the conversion to avoid audible noise dithering.


  10. #10
    Senior Member Bruce A. Richardson's Avatar
    Join Date
    Sep 1999
    Dallas, Texas

    Re: Creating samples: To normalize or not to normalize.

    Originally posted by SCARBEE:
    If you have recorded 24 bit and need to convert to 16 bit, it is a good idea to normalize before the conversion to avoid audible noise dithering.

    <font size=\"2\" face=\"Verdana, Arial\">Yes, it is.

    Let me clear up some misconceptions about normalizing.

    First, yes, it is a destructive edit, and on that level, you are indeed \"changing\" the sound.

    However, the bad onus once applied to normalizing is really moot. All of the modern mainstream editors process audio in 32-bit floating point before re-writing to whatever bit depth file you\'re using. Way back (well, six or seven years, maybe), some audio editors worked in whatever bit-depth the file was originally recorded. At that time, doing a normalize operation on a 16-bit file could indeed cause some signal degradation--especially after a series of edits.

    Today, that is not the case, and a normalize operation is going to be imperceptible for all practical purposes. However, if you perform many overall edits, you DO introduce a dithering step at every one--this CAN become quite perceptible, especially if your dithering algorithm is also performing noise shaping (i.e., Waves, Apogee, and most other high end dithering schemes). The cleanest way is to use a nondestructive editor (in other words, always keeping your original file and writing your edited file to a new filename). That means you\'re never introducing more than one dithering stage. Experiment freely, make note of every process\'s settings, then when you\'re ready to \"master\" your samples, set them up as a batch and run all the processes at one time.

    To the original question, though:

    YES. Normalize your samples. Here\'s why: You cannot set them up properly for wind/expression control if you don\'t. Wind controllers use attenuation to do their trick. If your sample itself is already attenuated, that is, soft samples are \"soft files,\" then a wind controller will, in effect, give double attenuation.

    GOS is an exception here, because it crossfades one sample into another, and that is what you link the wind controller\'s output into, rather than attenuation. But solo libraries, and most other cases, use attenuation.

    So--Normalize your samples, then use the built-in attenuation in the GigaStudio editor to knock their level back down to a natural level again. What you\'ll find helpful is making some files of the original dyanamic spread sampled, so you can use those comparatively to \"re-attenuate\" your instrument to a natural dynamic range.

    This may seem like a very anti-purist approach, but recording for samplers and recording for final product are two very different disciplines. In the end, you want your sample recordings to give the most flexibility possible in the programming stage, and normalizing gives you that flexibility.

    That\'s my take, anyway. I have spent plenty of time grumbling under my breath about non-normalized samples when I try to use my wind controller...whatever \"purity\" was gained in NOT normalizing was totally lost on my purposes as an end user.

    Of course, TASCAM could help us out by allowing per sample amplification as well as attenuation in the editor--then we\'d be able to boost a sample\'s level after the fact. And of course, if the sample producer doesn\'t use TASCAM\'s compression scheme, the end user can normalize and re-map the samples himself.

    This, to me, is a HUGE argument against using the compression scheme, and for that matter, any sort of copy protection methodology that limits the end user\'s ability to edit samples on the sample level. I understand why people do it, but that\'s a decision I wouldn\'t make as a designer. I have told the people at Nemesys/TASCAM that it is absolutely imperative to allow users to decompress these \"compressed\" libraries...and I hope others will join me in demanding it. If I cannot edit samples on the sample level, then it really hurts my ability to personalize libraries for the sound I\'d like them to have--which is very often a different sound that provided out of the box.

    Didn\'t mean to get into that, but I hear a lot of rumbling about copy protection, and this is yet another area where copy protection can really damage the end user\'s ability to work with the libraries he purchases. In short, sample producers should never make a decision that inconveniences paying users in order to somehow \"hurt\" pirates. Pirates never get hurt, and they always pirate what they want. Their money ain\'t on the table, so don\'t waste time and damage reputation over it!! Take care of the legitimate customer over all else, and the rest will take care of itself.

Go Back to forum

Tags for this Thread


Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts