Syntrillium Backup Forums :: View topic

SteveK50 · Posted - Sun Aug 05, 2001 6:20 pm

I'm looking at the latest demo and I've a question now.

I've dozens of music files, from various sources, and the volume varies. Is there a way to NORMALIZE every music file to a CONSISTENT volume level??

Thanks, Steve

rockindel1 · Posted - Sun Aug 05, 2001 10:31 pm

You can pick a normalization level( I use 95% ALL THE TIME) but the music may not have a consistant volume "sound" due to differing bass levels, if compression was used on the previous recording. Just like TV commercials "seem" louder due to compression they are actually at the same volume level,so the best answer is ...welll kind of, use your ears, open all songs on different windows and compare them.
ps also beware of peaks!!! when normalizing some songs will have a large cymbal crash or bass boom once and that will throw your whole normalization out,
good luck

jonrose · Location: USA

'YOUNGLOVE"S RMS NORMALIZATION FOUND!!!'

SteveK50 · Posted - Mon Aug 06, 2001 9:33 am

Thanks for both your responses; can't wait to get home to see if I can try things out and try to understand it all (audio dummy here).

I will clarify some... I've downloaded various MP3s and I guess each person, who created them in the first place, did them all different. Hence, I've some louder than others which shows up in the CE sine-wave... and I tried NORMALIZE on some last night but I just couldn't get them to (kind of) match up. I'd hate to burn a CD and have to keep raising/lowering the volume for each song

younglove · Posted - Mon Aug 06, 2001 11:47 am

Also, check out 2Bdecided's "Replay Level
calibration proposal" at
http://www.replaygain.org

This takes into account the Fletcher-Munson
curves (loudness of diferent frequencies).

An implementation of this to give a suggested
level change (an MS-DOS program:
"wavgain" for wave files) is to be found at
http://www.geocities.com/snelgman

beetle · Location: USA

This reoccuring question screams out for an RMS equalization function in Cool Edit!

Graeme · Member Location: Spain

alofoz · Location: Australia

Agreed (Beetle)

Agreed (Graeme)

But your ears must still be the final judge. RMS equalisation can still sound like the levels are different, depending on the type of music.

Cheers,
Alan

2Bdecided · Posted - Wed Aug 08, 2001 3:23 am

If syntrillium want to borrow the idea, they're very welcome to implement my Replay Gain Calculation in CE.

So come on Synt - there's a function for you, already designed and written, available for free!

Cheers,
David.
http://www.replaygain.org/

Graeme · Member Location: Spain

beetle · Location: USA

I find that the mathmatical calculations work just fine, and my ears can hear the cosistency in volume.

regger6 · Posted - Sat Sep 01, 2001 1:34 pm

One way that might help is by grouping songs that came from the same source (like from a CD you ripped) then using cool edit scripts with pause at dialogues checked to run a script using amplify set at 99 or what ever. You need pause at dialogues so you can hit recalculate for each song. If you just run the script without hitting recalculate, it'll use the same settings which might cause clipping if too high. This may work on mp3s after converting them to waves, or it may not work very well because of big peaks in the wave form. When this happens, you won't get much volume gain since the peak hits the top quickly. To fix this use hard limiting. Say if you know an entire CD is too low, find a hard limiting level that seems to boost the volume enough, then record a script and wha la, all the songs are louder.

beetle · Location: USA

Aaaah! Still too much trouble! Right now I use SF batch converter for this. They fixed all the bugs (finally).

Graeme · Member Location: Spain

beetle · Location: USA

regger6 · Posted - Wed Sep 05, 2001 10:14 pm

I agree, you have to be careful using hard limiting. But for a low volume CD it seems to me you can get away with 3db or so without much quality difference. I've been trying to get 2500 songs at reasonably consistent volumes. If low volumes are better, are you guys saying to lower all the rest of the volumes? It seems easier to me to boost the extremely low ones and cut the extremely high ones.
Rock on!

beetle · Location: USA

Again, regger6,

It's probably a good idea to either lower all of the files or set them to an average RMS. Making everything louder can destroy the dynamics.

Mr. Grynch · Posted - Fri Sep 14, 2001 5:17 pm

I've wrestled with this too, and the key appears to be in the RMS values. Here's what I do...

- Load in a .wav file
- Get statistics
- Look at Max RMS
- If it's between 4-7, I leave it alone
- If not, I add the left and right values together and divide by 2 to get an average value between the two channels
- I subtract 5 from the result. I find that 5 is a good median value
- I then AMPLIFY the waveform by that value
- I then normalize the waveform to 98%, you may want to use a lower value

I find this works really well in all cases, except for live recordings. Since the original editing for this waveforms are more erratic, I perform the same procedure, but instead use the AVG RMS and subtract 12 from the average, instead of 5

Please, let me know if this works for you, or if you have a better method. Obviously a method of normalizing based on RMS would be SO MUCH EASIER...<hint>

regger6 · Posted - Mon Nov 12, 2001 2:31 pm

Thanks beatle,
I stand corrected. The music sounds better bringing down loud tracks towards the quieter ones. This will definitely come in handy when producing my own heavy metal stuff! Thankfully, just to listen to mp3's on my computer, I finally figured out how to make winamp automatically adjust the volumes.
Rich

AusRob · Location: Australia

Another angle to this discussion....?
... from my earlier years as an acoustics engineer, I recall the concept of "L10" or "L50", etc, where the sound level is exceeded for 10% or 50% of the time.... and "Leq" represented an "average" level, which I guess is similar to RMS level. Useful for describing transient noise levels, such as traffic or impulsive industrial noise. Is there a similar concept in audio engineering? How good a descriptor of audio signal loudness would (say) an L10 level be? Could it be that such a scheme could provide some consistency in apparent loudness?

Best Regards, ROB

SteveG · Location: United Kingdom

That's an interesting idea, Rob. I did a stint putting ticks in boxes whilst trying not to get run over as well! Yeah, we did it the hard way. Then they told us about the B&K counter and using a tape recording!

But surely the narrower and more consistent the dynamic range, the harder it would be to determine the L values?

I recall from psychoacoustics doing calculations of perceived noise (PNdB) involving noy values, and there was an antiquated Noise and Number Index (NNI) for aircraft noise. I'm sure that what David's doing with his replay gain calculation is going to be conceptually similar to some of these measurements. But I certainly wasn't aware of any similar work in the music area until he announced the results of his research.

Steve

AusRob · Location: Australia

Steve, the organization I worked for fortunately had (for the time) pretty flashy B&K analysis gear (the counter with strip hard-copy recorder), however it was still a laborious process... many hours in the lab, listening to tape recordings of traffic or industrial noise, watching the printout in real-time, and editing out any extraneous influences such as coughs, sneezes, and other bodily noises....
Many were the times I napped while listening to these recordings.

My work at the time focussed on determining the best way to "standardize" measurement of both traffic and impulsive industrial noise, and I just thought that similar priciples may be usable in audio engineering. With a narrower dynamic range, less "normalization" would be required to bring a signal to a given level, and the converse would apply for signals with wider dynamic range...?? Also, I presume that the algorithm for an "Lxx" level may be pretty simple??

Cheers, ROB

AndyH · Posted - Mon Nov 12, 2001 5:23 pm

You might want to look at the shareware program Volume Balancer at
http://homepages.nildram.co.uk/~abcomp/
It won't make at the potential concerns that people have expressed here go away, but it seems designed to make things easy, while retaining as much qualtity as possible.

I have not wanted to do this, so I haven't tried it, but I know that another program by the same author, WaveRepair, is very well done.

SteveG · Location: United Kingdom

2Bdecided · Posted - Tue Nov 13, 2001 6:40 am

AusRob,

I'm doing something similar - not because I had the brains to figure it out for myself, but because it's mentioned in one of Zwicker's many papers on the topic. HE developed one of the "accepted" noise measurements in the late 1970s.

I'm effectively using L05 - which sounds very high, but it seems to work.

The psychoacoustics of judging the loudness of short sounds is very well worked out. The ways of combining all these loudness judgements into some kind of overall loudness (Leq? maybe not) still seem to be quite primative - maybe because this is a difficult take to get human listeners to agree on!

I've missed a lot of the pyschoacoustic principles out of the replay gain proposal because the calculation needs to be fast. The filtering into critical bands and smearing of the time-domain envelope would make things more accurate, but since the calculation of Leq is such a (shall we say) bodged effort, all the extra psychoacoustic knowledge upstream wouldn't really help for most real music signals.

To answer your question - yes, the L10 (or the L05 I have chosen to use) does provide a good consistency in loudness - even between tracks of almost zero dynamic range, and tracks with a huge dynamic range (though ofcourse it's the "middle" of the loudness range which is matched - despite the choice of L05).

Can you give me any pointers to how you calculated the overall perceived loudness? Did you just use an Lxx, or the Leq? Or were there any other tricks you picked up which would be useful in what I'm trying to do?

Cheers,
David.
http://www.David.Robinson.org/
P.S. - this isn't my research, this was a tangent to my research that I couldn't resist following!

AusRob · Location: Australia

David,
Thanks for your response... this is very interesting. I have specialised in thermal engineering for the last 23 years, and have had very little involvement in acoustics in that time, unfortunately, so I'm afraid I would be unable to contribute much further on an academic level. I share my office with one of my old acostics "cohorts", so I'll ask him for his opinion or whether he can point you in the right direction for some further advice.

Best Regards, ROB

SteveG · Location: United Kingdom

I've now had some time to think about it, and I can certainly see the attraction of using L05. I presume that for speed, you are basing your calculation on timed measurements of instantaneous values rather than an FFT based approach? I'm still quite impressed that this would give consistent results across a wide range of music types - this must mean that there's more consistency across musical genres than I thought!

Steve

Postscript - I checked it out on the website, and found the answer! Should have looked in the first place...(slap on the wrist)

Edited by - SteveG on 11/13/2001 5:22:24 PM

sk · Location: USA

Thanks Andy H for your suggestion. I went to that site and downloaded Volume Balancer and bought it ($15) after using it for about 3 minutes! It really does seem to do what it says, with a minimum of fuss, with decent documentation, and without needing to go back to college to earn an engineering degree to use it. :-)
(No ill-intent towards engineers!! It's just that I'm obviously not one, and appreciate programs that incorporate the necessary fundamentals in such a way that the program just works and does what it says without needing to necessarily understand how or why. I think a good example of this principle is an automobile - I am not a mechanic and have no idea how or why my car works the way it does. But that doesn't mean I'm not a good, responsible driver.)
sk

sk · Location: USA

To Andy H again:
Once again, thanks.
Really thanks!
I just finished balancing 22 songs that I've been fiddling with for about 3 months now. I had access to the RMS in Sound Forge that Beetle keeps wishing CE would add so he could just stick with CE completely, and it got me pretty close to the target range. But Volume Balancer just operates on a whole different level. I still needed to fine tune a couple of songs after running it (which took a grand total of about 15 minutes to 'prepare' and then 'balance' all 22 songs), and I did choose the 'foolproof' method of choosing to target the lowest volume song and bring everything down to that level rather than the other way around. But I avoided any compression and that seemed to allow the music to 'breathe' a bit. It was a very satisfying experience listening to 22 songs in a row that definitely flowed together as a whole. (With the few exceptions that I mentioned that needed a little final tweaking.) The program had, to my ear, a definite professional fit and finish that I clearly was unable to achieve just using the RMS values. It took me a while to really listen to realize how much cleaner the sound was. At first I missed that 'punch'. But I didn't miss the raspy harshness that often accompanies the 'punch'. And once my ears adjusted I was hearing parts of the song that I was not hearing before when they were set higher. It was a very satisfying experience, and what makes it all the more amazing to me is that the whole program costs 15 bucks! Keep wailing away with the Peaks and RMS, Beet. I'm switching to Volume Balancer. lol.
sk