How to equalise loudness across a number of voice recordings

Discussion forum for Amadeus users

Moderator: Martin Hairer

Post Reply
ogyen
Posts: 15
Joined: Sun Sep 02, 2012 10:40 am

How to equalise loudness across a number of voice recordings

Post by ogyen »

Hi

You helpful people gave great advice when i started archiving old cassetes of buddhist teachings...about ten years ago!

I have just recorded a new translation of the Dhammapada (a major buddhist text) over a period of a couple of months.

There are 26 files short (5mins or so) and i would like them to have the same sense of loudness so people dont have to make adjustments if they listen to one after another.
I would like the listening to be a pleasure...it will be offered for free.

I have converted one file to mp3...and will have a go at doing the rest in the batch processor.

Looking at the answers to the one similar question... normalising to -1 peak is suggested. Another suggeston is to normalise to -8db max or -10db.

or I could do both! one after the other...
What do you suggest?

Maybe the sample statistics below will be helpful?

The wave stats for the MP3... which may have gone through izotope before conversion (I can't remember now)
are
Peak amplitude -8
DC offset 0
minimum RMS power -inf dB
average RMS power -35.2dB
maximum RMS power -19.8 dB

For the first .wav file (which corresponds to the MP3 above)
Peak amplitude is -8
DC offset 0
minimum RMS power -97.5 dB
average RMS power -35.2dB
maximum RMS power -19.7

For the final .wav file

Peak amplitude -8.6
DC offset zero
Minimum RMS power left channel -infdB right channel -74.2 dB
average RMS power -35.2 dB
maximum RMS power -20.9 DB

another I selected at random because it looked a little 'thin' has
peak amplitude -3.9
minimum RMS -80
average RMS -37.
maximum RMS -17.3

any advice would be much appreciated.

wishing you good health and Happy Easter
Wendy

philxm
Posts: 128
Joined: Sun Nov 12, 2006 6:55 pm

How to equalise loudness across a number of voice recordings

Post by philxm »

A technique I use in similar situations helps me quickly gain a sense of which files in a batch require boosting and vice versa. It’s also very simple:
Use Sound menu > Join Files to open all the files as one long file. (Perhaps best to do this close to the end of your processing process.)

When you visually examine the entire resulting file in a single window, you are liable to notice some of the included files looking less powerful than the average, and perhaps some others looking more so. From this same view you can also quickly play bits of those, and compare them to more average-appearing files.

You can also make adjustments to individual files this way, and, again, quickly compare results. When done, if you don’t make any changes you can simply close the file without saving, but if you did make changes use Sound menu > Split According To Markers, which will save new versions of each file in the batch.
On Apr 11, 2020, at 3:50 PM, ogyen <forum2mail@hairersoft.com (forum2mail@hairersoft.com)> wrote:
Hi You helpful people gave great advice when i started archiving old cassetes of buddhist teachings...about ten years ago!I have just recorded a new translation of the Dhammapada (a major buddhist text) over a period of a couple of months. There are 26 files short (5mins or so) and i would like them to have the same sense of loudness so people dont have to make adjustments if they listen to one after another.I would like the listening to be a pleasure...it will be offered for free.I have converted one file to mp3...and will have a go at doing the rest in the batch processor.Looking at the answers to the one similar question... normalising to -1 peak is suggested. Another suggeston is to normalise to -8db max or -10db.or I could do both! one after the other...What do you suggest? Maybe the sample statistics below will be helpful?The wave stats for the MP3... which may have gone through izotope before conversion (I can't remember now) arePeak amplitude -8 DC offset 0 minimum RMS power -inf dB average RMS power -35.2dB maximum RMS power -19.8 dB For the first .wav file (which corresponds to the MP3 above) Peak amplitude is -8 DC offset 0 minimum RMS power -97.5 dB average RMS power -35.2dB maximum RMS power -19.7 For the final .wav filePeak amplitude -8.6 DC offset zero Minimum RMS power left channel -infdB right channel -74.2 dB average RMS power -35.2 dB maximum RMS power -20.9 DBanother I selected at random because it looked a little 'thin' has peak amplitude -3.9 minimum RMS -80 average RMS -37. maximum RMS -17.3any advice would be much appreciated.wishing you good health and Happy Easter Wendy

User avatar
Jim Edgar
Posts: 118
Joined: Mon Mar 01, 2010 7:20 am
Contact:

How to equalise loudness across a number of voice recordings

Post by Jim Edgar »

RMS is a better way to balance.

Peak is simply the loudest single cycle (wave) in the file. One loud click could throw things off if you are normalizing by Peak.


Using stock plugins, you could use the limiter and then normalize down to a reasonable level.


Typical RMS targets for audiobooks are -18 to -23dB. That assumes clean audio.


Happy to answer more on this topic but away from computers for most of today.


Jim
JustAskJimVO.studio

--sent from some phone






On Apr 11, 2020, at 12:58 PM, "pm@philxmilstein.com (pm@philxmilstein.com)" <pm@philxmilstein.com (pm@philxmilstein.com)> wrote:

A technique I use in similar situations helps me quickly gain a sense of which files in a batch require boosting and vice versa. It’s also very simple:
Use Sound menu > Join Files to open all the files as one long file. (Perhaps best to do this close to the end of your processing process.)

When you visually examine the entire resulting file in a single window, you are liable to notice some of the included files looking less powerful than the average, and perhaps some others looking more so. From this same view you can also quickly play bits of those, and compare them to more average-appearing files.

You can also make adjustments to individual files this way, and, again, quickly compare results. When done, if you don’t make any changes you can simply close the file without saving, but if you did make changes use Sound menu > Split According To Markers, which will save new versions of each file in the batch.
On Apr 11, 2020, at 3:50 PM, ogyen <forum2mail@hairersoft.com (forum2mail@hairersoft.com)> wrote:
Hi You helpful people gave great advice when i started archiving old cassetes of buddhist teachings...about ten years ago!I have just recorded a new translation of the Dhammapada (a major buddhist text) over a period of a couple of months. There are 26 files short (5mins or so) and i would like them to have the same sense of loudness so people dont have to make adjustments if they listen to one after another.I would like the listening to be a pleasure...it will be offered for free.I have converted one file to mp3...and will have a go at doing the rest in the batch processor.Looking at the answers to the one similar question... normalising to -1 peak is suggested. Another suggeston is to normalise to -8db max or -10db.or I could do both! one after the other...What do you suggest? Maybe the sample statistics below will be helpful?The wave stats for the MP3... which may have gone through izotope before conversion (I can't remember now) arePeak amplitude -8 DC offset 0 minimum RMS power -inf dB average RMS power -35.2dB maximum RMS power -19.8 dB For the first .wav file (which corresponds to the MP3 above) Peak amplitude is -8 DC offset 0 minimum RMS power -97.5 dB average RMS power -35.2dB maximum RMS power -19.7 For the final .wav filePeak amplitude -8.6 DC offset zero Minimum RMS power left channel -infdB right channel -74.2 dB average RMS power -35.2 dB maximum RMS power -20.9 DBanother I selected at random because it looked a little 'thin' has peak amplitude -3.9 minimum RMS -80 average RMS -37. maximum RMS -17.3any advice would be much appreciated.wishing you good health and Happy Easter Wendy


CDJonah
Posts: 119
Joined: Sat Nov 11, 2006 12:10 am

How to equalise loudness across a number of voice recordings

Post by CDJonah »

When I was creating CDs, I used to use an app, whose name escapes me at present, that would calculate the average frequency -weighted power for each item. I would then create an Excel file for each file with the maximum signal and the average frequency weighted power and normalize everything to the same frequency-weighted power using the maximum to make define what the maximum would be. This would come pretty close and then I would listen and adjust for a final.

I was doing classical vocal music over a wide range of periods so that if you would normalize a Vivaldi and say Schubert to the same maximum, you quickly found that the Vivaldi felt like it was hammering you. I would think that average RMS could substitute for the app I used to use.

Chuck

On 4/11/20, 2:50 PM, "forum_list on behalf of ogyen" <forum_list-bounces@hairersoft.com on behalf of forum2mail@hairersoft.com> wrote:

Hi

You helpful people gave great advice when i started archiving old cassetes of buddhist teachings...about ten years ago!

I have just recorded a new translation of the Dhammapada (a major buddhist text) over a period of a couple of months.

There are 26 files short (5mins or so) and i would like them to have the same sense of loudness so people dont have to make adjustments if they listen to one after another.
I would like the listening to be a pleasure...it will be offered for free.

I have converted one file to mp3...and will have a go at doing the rest in the batch processor.

Looking at the answers to the one similar question... normalising to -1 peak is suggested. Another suggeston is to normalise to -8db max or -10db.

or I could do both! one after the other...
What do you suggest?

Maybe the sample statistics below will be helpful?

The wave stats for the MP3... which may have gone through izotope before conversion (I can't remember now)
are
Peak amplitude -8
DC offset 0
minimum RMS power -inf dB
average RMS power -35.2dB
maximum RMS power -19.8 dB

For the first .wav file (which corresponds to the MP3 above)
Peak amplitude is -8
DC offset 0
minimum RMS power -97.5 dB
average RMS power -35.2dB
maximum RMS power -19.7

For the final .wav file

Peak amplitude -8.6
DC offset zero
Minimum RMS power left channel -infdB right channel -74.2 dB
average RMS power -35.2 dB
maximum RMS power -20.9 DB

another I selected at random because it looked a little 'thin' has
peak amplitude -3.9
minimum RMS -80
average RMS -37.
maximum RMS -17.3

any advice would be much appreciated.

wishing you good health and Happy Easter
Wendy

User avatar
Jim Edgar
Posts: 118
Joined: Mon Mar 01, 2010 7:20 am
Contact:

How to equalise loudness across a number of voice recordings

Post by Jim Edgar »

LUFS / RMS are the current standards, depending upon your field. 

Youlean Loudness Meter is a good free AU plugin
https://youlean.co/youlean-loudness-meter/





Jim Edgar

VO: JimEdgarVoices.com | @jimedgarvoices | Source-Connect - jimedgarvoices


For audio/studio help: JustAskJimVO.studio | Schedule a session


















On Sat, Apr 11, 2020 at 4:58 PM Charles Jonah <chuck@charlesjonah.org (chuck@charlesjonah.org)> wrote:
When I was creating CDs, I used to use an app, whose name escapes me at present, that would calculate the average frequency -weighted power for each  item. I would then create an Excel file for each file with the maximum signal and the average frequency weighted power and normalize everything to the same frequency-weighted power using the maximum to make define what the maximum would be. This would come pretty close and then I would listen and adjust for a final.

I was doing classical vocal music over a wide range of periods so that if you would normalize a Vivaldi and say Schubert to the same maximum, you quickly found that the Vivaldi felt like it was hammering you. I would think that average RMS could substitute for the app I used to use.

Chuck

On 4/11/20, 2:50 PM, "forum_list on behalf of ogyen" <forum_list-bounces@hairersoft.com (forum_list-bounces@hairersoft.com) on behalf of forum2mail@hairersoft.com (forum2mail@hairersoft.com)> wrote:

    Hi

    You helpful people gave great advice when i started archiving old cassetes of buddhist teachings...about ten years ago!

    I have just recorded a new translation of the Dhammapada (a major buddhist text) over a period of a couple of months.

    There are 26 files short (5mins or so)   and i would like them to have the same sense of loudness so people dont have to make adjustments if they listen to one after another.
    I would like the listening to be a pleasure...it will be offered for free.

    I have converted one file to mp3...and will have a go at doing the rest in the batch processor.

    Looking at the answers to the one similar question... normalising to -1 peak is suggested. Another suggeston is to normalise to  -8db max or -10db.

    or I could do both! one after the other...
    What do you suggest?

     Maybe  the sample statistics below will be helpful?

    The wave stats for the MP3... which may have gone through izotope before conversion (I can't remember now)
    are
    Peak amplitude -8
     DC offset 0
     minimum RMS power -inf dB
     average RMS power -35.2dB
     maximum RMS power -19.8 dB

     For the first .wav file (which corresponds to the MP3 above)
     Peak amplitude is -8
     DC offset 0
     minimum RMS power -97.5 dB
     average RMS power   -35.2dB
     maximum RMS power -19.7

     For the final .wav file

    Peak amplitude -8.6 
    DC offset zero 
    Minimum RMS power left channel -infdB     right channel -74.2 dB
     average RMS power  -35.2 dB
     maximum RMS power -20.9 DB

    another I selected at random because it looked a little 'thin' has 
    peak amplitude -3.9
      minimum RMS -80
     average RMS -37.
    maximum RMS -17.3

    any advice would be much appreciated.

    wishing you good health and Happy Easter
     Wendy








ogyen
Posts: 15
Joined: Sun Sep 02, 2012 10:40 am

Many thanks... and more questions!

Post by ogyen »

Well thank you so much for the speedy replies and the intelligent thought behind them.
I am sure i will be able to make a good job of this with your help.
I was busy yesterday with web-site stuff, students queries -I teach dharma (Dzogchen and other approches) and help with peoples anxieties for free- and making my own face-mask!

So firstly it ocurred to me to ask if any of you would like to collaborate on this project and do this work for the benefit of others?
Its a long shot and you are no doubt, both here and other places helping others for free.
You would be saving me some time which would go to teaching/practice/maybe even hoovering ...which hasn't been done for months!

Phils good idea gave me another one ...that it would be good in any case to have – as well as 26 short recordings... to reflect on each (as a nugget!) – a full legth recording of 2hrs or so which could be made by joining the files. So thanks for that...and for the 'by eye' way of evaluating noticable changes in file 'strength'. Would you suggest acheiving homogeneity by altering this by changing the gain of the file which was out of kilter to visually match the others ?

Jims advice to work to RMS -18 to -23 RMS (assuming this is peak RMS Jim?) for audio books...is very helpful. Almost all are within that range, one i noticed is -24 others are mostly -19. Should I do some more work with the loudness plug-in Jim kindly sent...thanks you for that...and then some normalising?
I'd really appreciate some detailed help with this if i'm going to try this.

Chuck's experience (thank you Chuck) suggests using average RMS as the value to match across the board. Is it the gain which i would alter in each file to aceive this...to match that of a file which sounds good? Hoping that the peak RMS still fits within Jims parameters?
The normalising processor in Amadeaus, as i understand it just works with peak RMS...so maybe I could alter the gain in each file manually to get average RMS , after i have altered the Peak RMS to say -20%...
Reading that i think it sounds like garbage...so perhaps someone could explain a sensible process to me?


As you can see i'm a bit out of my depth here...but will swim with your capable advice/help....or get an ice-cream (those were the days!) if someone would like to do this for me?

Thanks again, i'll look forward to hearing from you : )
best wishes
wendy

Jonah, Charles D. via ...

How to equalise loudness across a number of voice recordings

Post by Jonah, Charles D. via ... »

See below

Chuck's experience (thank you Chuck) suggests using average RMS as the value to match across the board. Is it the gain which i would alter in each file to aceive this...to match that of a file which sounds good? Hoping that the peak RMS still fits within Jims parameters?
The normalising processor in Amadeaus, as i understand it just works with peak RMS...so maybe I could alter the gain in each file manually to get average RMS , after i have altered the Peak RMS to say -20%...
================================
Let us give an example:
RMS average peak peak at -15
-18 -7 -4
-23 -5 +3
-15 -10 -10
So to take them all to -15 dB average
The first sample would be 3 dB amp, second would be 8 dB and the 3rd would be 0 dB but clearly that wont work, the peak would be for the second sample would be +3dB -- not doing that. So we would take the second sample to -18 giving a peak of 0, the second t0 0 and the 3rd to -13 (-3dB amplification)
Do not use peak RMS, you use absolute max peak

Chuck

User avatar
Jim Edgar
Posts: 118
Joined: Mon Mar 01, 2010 7:20 am
Contact:

How to equalise loudness across a number of voice recordings

Post by Jim Edgar »

Most Normalize processes work with either Peak (not Peak RMS) which is the loudest single wave in the file.  They can also be set to Normalize by RMS (in AP, it actually says it does so with the Maximal RMS - most use Average).  In either case if you change one, it changes the other by the same amount.  E.G. if you have an RMS of -24 dB and a Peak of -6 dB, then Normalizing by RMS to -20 dB would raise the peak to -2 dB (because everything went "up" by 4 dB.) So you have to be mindful that increasing by RMS does not push the peaks past 0 dB (causing distortions). 

Now, honestly, I don't use AP for this these days. So, I'm a bit confused by the Normalize effect in AP. 


It gives me the option of Normalizing to Maximal Peak (which does make sense) or by Maximal RMS (as opposed to "Average RMS"). When I Normalize to -23 RMS (Maximal), it actually generates a file with a Maximal RMS Peak of -21 according to waveform statistics.  

Martin - by any chance is the Normalize process using older AES Standard, while the Waveform Statistics is using AES 2017?
If so, audiobooks are evaluated with the older AES standard still, so an option to control that might be helpful.


Back to the process:
I would use a Limiter to boost the RMS and control the Peak
The AUPeakLimiter will do this. 
Set a moderate PreGain - depending upon your source it may be in the +4 to +12 range
Apply
This will boost the RMS while constraining the Peak at -0 dB
Then use Normalize to bring the overall volume down.


There are 3rd Party Tools (Loudmax from Thomas Mundt being a very nice, non-coloring effect) which let you control the Peak and Threshold (the "PreGain" in AU's tool) which would eliminate the need to Normalize afterward.


Hope that helps a bit.


RMS = Root Mean Squared 
LUFS = Loudness Units Full Scale
both of the above are overall weighted averages designed to give realistic (i.e. the way we perceive volume) references for how loud something is. Our hearing doesn't necessarily track peak cycles and we also tend to be more to certain frequencies in the frequency range. 




Jim EdgarVO: JimEdgarVoices.com | @jimedgarvoices | Source-Connect - jimedgarvoices


For audio/studio help: JustAskJimVO.studio | Schedule a session














On Mon, Apr 13, 2020 at 4:41 AM ogyen <forum2mail@hairersoft.com (forum2mail@hairersoft.com)> wrote:
Well thank you so much for the speedy replies and the intelligent thought behind them.
I am sure i will be able to make a good job of this with your help.
I was busy yesterday with web-site stuff, students  queries -I teach dharma (Dzogchen and other approches) and help with peoples anxieties for free- and making my own face-mask!

So firstly it ocurred to me to ask if any of you would like to collaborate on this project and do this work for the benefit of others?
Its a long shot and you are no doubt, both here and other places helping others for free.
You would be saving me some time which would go to teaching/practice/maybe even hoovering ...which hasn't been done for months!

Phils good idea gave me another one ...that it would be good in any case to have – as well as  26 short recordings... to reflect on each (as a nugget!) – a full legth recording of 2hrs or so which could be made by joining the files. So thanks for that...and for the 'by eye' way of evaluating noticable changes in file 'strength'. Would you suggest acheiving homogeneity by altering this by changing the gain of the file which was out of kilter to visually match the others ?

Jims advice to work to RMS -18 to -23 RMS (assuming this is peak RMS Jim?) for audio books...is very helpful. Almost all are within that range, one i noticed is -24 others are mostly -19. Should I do some more work with the loudness plug-in Jim kindly sent...thanks you for that...and then some normalising?
I'd really appreciate some detailed help with this if i'm going to try this.

Chuck's  experience (thank you Chuck) suggests using average RMS as the value to match across the board. Is it the gain which i would alter in each file to aceive this...to match that of a file which sounds good? Hoping that the peak RMS still fits within Jims parameters?
 The normalising processor in Amadeaus, as i understand it just works with peak RMS...so maybe I could alter the gain in each file manually to get average RMS , after i have altered the Peak RMS to say -20%...
Reading that i think it sounds like garbage...so perhaps someone could explain a sensible process to me?


As you can see i'm a bit out of my depth here...but will swim with your capable advice/help....or get an ice-cream (those were the days!) if someone would like to do this for me?

Thanks again, i'll look forward to hearing from you : )
best wishes
wendy




philxm
Posts: 128
Joined: Sun Nov 12, 2006 6:55 pm

How to equalise loudness across a number of voice recordings

Post by philxm »

Phil's good idea gave me another one ...that it would be good in any case to have – as well as 26 short recordings... to reflect on each (as a nugget!) – a full legth recording of 2hrs or so which could be made by joining the files. So thanks for that...and for the 'by eye' way of evaluating noticable changes in file 'strength'. Would you suggest achieving homogeneity by altering this by changing the gain of the file which was out of kilter to visually match the others ?
To be clear, the benefit of my method (to group the files into a single long file) is to use the visual image as a quick way to judge which sub-files appear to be too quiet or too loud. It’s also useful in that one can both see and listen quickly to portions of the included sub-files, but since we’re working with audio, the ear will always be a better judge than the eye.

Frankly, my knowledge of audio processing isn’t sophisticated enough to make sense of RMS, but since my needs are equally modest the method I described is perfectly suitable to me.

ogyen
Posts: 15
Joined: Sun Sep 02, 2012 10:40 am

Post by ogyen »

Hi to all of you.. and many thanks for your replies

I had a birthday and a mini internet-holiday to celebrate!

You have offered me a number of routes to try from simple to comlex... and also links to software which will be very helpful and are much appreciated.
[ Yes, thaks Phil... understood :) ]

I'm sure they all have their merits and I'll enjoy trying them out and see how i go...

...and if any one would like a link to the final result just let me know (it's free, no strings!)
Though on reflection i'm not sure that I can post it here...
Perhaps there is a way?
Usually...when there is a will there is one.... :)

Thanks again,

and all good wishes for your ease, health, and happiness

wendy

User avatar
Martin Hairer
Site Admin
Posts: 1975
Joined: Wed Nov 08, 2006 11:49 am
Contact:

How to equalise loudness across a number of voice recordings

Post by Martin Hairer »

Though on reflection i'm not sure that I can post it here...
Perhaps there is a way?
You are welcome to post a Dropbox link (or similar), provided of course that you do have the right to distribute the file it links to. Best,

Martin

--
HairerSoft
http://www.hairersoft.com/

ogyen
Posts: 15
Joined: Sun Sep 02, 2012 10:40 am

Post by ogyen »

Thanks for that Martin...that's kind!

So yes...please let do me me know if you'd like a link to hear me reading this up-to-date, yet authentic, translation of the dammapada...and i'll send it to you when it's published.

It is a recent revision by James Low (my teacher) of his original translation, made in 1982, of a translation of the Pali text made by the Tibetan monk Gendun Choepal!

The original text is available for reference. I scanned it from James' copy, one of the few remaining copies of the book, which contains the verses in Pali Sansrit and Tibetan... with Hindi and English translations.

I teach within the same lineage and have authorisation to offer this...
Wisdom does not grow old and its value is priceless, yet how it is expressed is critical if it is to be useful...wx

Post Reply