src conversion from 96Khz to 44.1 - gives volume / peak change

[inlinenl]

hello ME's : sometimes when I'm mastering in the 24/96Khz domain ... ( music which benefits from this and using the weiss DS1 and EQ1 internal sample rate 96 Khz ) and I put my peak volume at -0.50 dB ( L2-output-volume ) .. I have the problem or question when converting to 24/44.1 Khz ...

.... when applying SRC conversion in wavelab 5.0 the peaks sometimes/always goes to 0.0 dB or just changing to another value .... resulting in a non-hearable click/over ...

I just lower the volume on the 24/96 khz file to go to an < 0.01 dB in the 24/44.1 khz domain ..
normally when working in 44.1 Khz I always aim to -0.20 dB or -0.30 dB

I just feel a bit stupid lowering the volume to -0.8 dB ... ( there goes my carefully build-up volume )

what is you're exp. and do other src/conversion programms act the same way ...
I'm using the internal src converter or the added plug-in ...

thanks in advance ...

[peeder]

The SRC will be a reconstructing filter. And as such, your new peaks are probably actually inter-sample peaks...you were actually generating voltages above 0dbFS, but they happened to fall between samples at the higher rate. With the new set of samples, you have hit some peaks more square-on.

Although it may be that your SRC is goosing the gain a bit or it is resonating or aliasing which it shouldn't.

Anyway if you have to use a PC try Saracon or R8 Brain Pro and see if they do the same thing. There's a free R8 brain to try too.

According to SRC Comparisons Wavelab 5 sucks at SRC.

[inlinenl]

thanks peeder, will try the r8brain ... curious if sarcon will act the same way ....
maybe george could tell us ...

[aivoryuk]

R8BRain won't solve the problem as you will still get overs, but on the pro version you can set the conversion so it lowers it automatically (but you still might be subjective to intersample peaks)

You then can use another peak limiter to raise the volume to what it was. Although i think this will change the sound of the master slightly

[MattGray]

This is a common problem & it is definitely intersample peaks from the SRC process. I've not found an SRC that will totally eliminate this occurance however there is a couple of approaches you can take. Kind of like 'pick your poison' & quite program dependent.

1. Limit to 0.0dbfs at 96kHz do the SRC to 44.1kHz then lower the final output to -0.3dbfs pre final dithering to 16bit. Ignoring the intersample peak overs...
2. Lower the output of the limiter to -0.3 or -0.5 & ignore the few overs that occur during the SRC to 44.1kHz. (as you are already doing)
3. Do your final limiting post SRC at 44.1kHz which won't sound quite as good but it will eliminate the intersample peaks in most cases. Here you are trading off 96kHz operation against intersample peak distortion, you have to decide if this is a worthwhile tradeoff.
4. Remove the limiter & simply use your A/D to capture at 24-44.1 clipping on the way in instead of limiting. Once captured assemble your master do your fades etc & then lower the final output to -0.3dbfs prior to the final 16bit dithering. I bet in most cases with the right A/D this will sound better then the same gain reduction with your L2. You need a good A/D for this & obviously you can't push it too far (peaks only!). The benefits are this.. You avoid an SRC process & therefore the intersample peak issue. This assumes you are using mostly analog processing or at least using the analog processing last in the chain.

It's up to you to choose the best sounding option for the track...

[arf]

Originally Posted by MattGray

The benefits are this.. You avoid an SRC process & therefore the intersample peak issue. This assumes you are using mostly analog processing or at least using the analog processing last in the chain.

Well, all modern ADCs digitize at oversampled rates and use on board real-time SRC to get down to the target rate. So it comes down to whether the SRC in the converter is preferable to SRC done externally with software or another digital device. That, of course, depends on what converters and/or software you have.

[bob katz]

Originally Posted by inlinenl

thanks peeder, will try the r8brain ... curious if sarcon will act the same way ....
maybe george could tell us ...

The Weiss SRC (and the hardware) has always incorporated an 0.2 (or possibly 0.3, I forget) dB gain drop to protect from intersample peaks. Saracon also has a lot to tell you about the modulation if you want to check it. But so far I haven't had any overload issues with Saracon.

However, if you are peak limiting extremely aggressively and/or pushing and using extreme equalization, I would be cautious about ANY SRC and consider measuring the output with an intersample peak meter and possibly redo even a few tenths of a dB lower. If not, you risk getting harsh distortion with many cheaper DACs. It's not unusual here when being over-aggressive (read: "smashed") to be quite shocked at how bad the results are on DSP-based home reproduction units like the Aiwa 3 piece systems. It's not nice to fool mother nature!

[bob katz]

Originally Posted by MattGray

This is a common problem & it is definitely intersample peaks from the SRC process. I've not found an SRC that will totally eliminate this occurance however there is a couple of approaches you can take. Kind of like 'pick your poison' & quite program dependent.

1. Limit to 0.0dbfs at 96kHz do the SRC to 44.1kHz then lower the final output to -0.3dbfs pre final dithering to 16bit. Ignoring the intersample peak overs...

I would only risk this if the SRC is set to produce a floating point file on its output, which gives you full options. Otherwise the distortion damage has already been done if the SRC's output distorts in the fixed point domain. Any time the signal crosses domains, intersample peaks and resultant audible distortion are possible, and the SRC is a "domain crosser". Not as damaging to the sound as an overloaded DAC or mp3 file, but still damaged.

BK

[arf]

Originally Posted by bob katz

However, if you are peak limiting extremely aggressively and/or pushing and using extreme equalization, I would be cautious about ANY SRC and consider measuring the output with an intersample peak meter and possibly redo even a few tenths of a dB lower.

Good advice, but can't intersample peaks be on the order of a few dB or more?

So why is there so much talk about lowering .2dB - what good is that going to do for over-limited/over-clipped induced distortion in cheap a consumer DAC?

Live by the clip, die by the clip!

[jdg]

mildly OT, but, this is why i prefer to playback at one rate, and capture at 44.1 on a 2nd daw.

before i spend a few hours listening to some SRC'd 96->44 vs played back 96-> captured at 44.1 files, to see if one sounds better in general to me... do ppl already have a preference for this?

i know that alot of us already use 2 daws for this reason...
just curious.

[bob katz]

Originally Posted by arf

Good advice, but can't intersample peaks be on the order of a few dB or more?

So why is there so much talk about lowering .2dB - what good is that going to do for over-limited/over-clipped induced distortion in cheap a consumer DAC?

Live by the clip, die by the clip!

Avast ye mateys! Well, actually, while theoretically intersample peaks could conceivably be a few dB above the digital level, in practicality this rarely occurs. So Weiss's choice of 0.2 dB usually is enough, with material that is not extremely aggressively processed. The thing is, even if you take that aggressively processed material down so an oversampled meter says it is not going over, it is not always enough to deal with mp3 conversion or cheap DACs.

Also remember that oversampled metering is only approximation of what MIGHT happen in a "typical" filtering situation. I now have 4 different oversample-capable meters and do see 0.1 to 0.2 dB difference between my TC MD4 (or limiter section) and my RME Digicheck when set to OVS and especially the PSP Xenon.

We should keep a list of which meters are now capable of oversampling:

PSP

TC 6000

RME Digicheck

SSL has a free download that can do it now, too.

BK

[plush2]

Bitter by Schwa is a VST meter plugin somewhat similar to the SSL one that will detect intersample peaks and oversample. It's free as well.

Stillwell Audio - It’s About The Sound » Bitter

[peeder]

Originally Posted by bob katz

We should keep a list of which meters are now capable of oversampling:

TL MasterMeter for Pro Tools....

[inlinenl]

intersample ??? the space between two samples ?? intersamples don't exist ....
intersamples are the samples in the higher sample rate which will be "summed/recalculated " as one in the lower rate .. giving "sometimes" an over ( simpel conclusion ?? ) .... thanks for clearing me on that

... it acctually happend with a very good produced/sounding dance production which had to go loud .... normally I keep these in de 44.1 range .. but I thought this production good live with a 96 start ...

normally 96 is mostly used in accoustic productions where there is no need for high RMS and level pushing ... it's based on source and not on result so there's no need to check for hours if it's sounding better than staying in the 44.1

when I master .. all is going in 1 run from source to processing to destination ... so the captured file will be non-processed further ... only src and dither and thats it ... I don't want to touch the captured file with any other tools ...

there for I just set the L2 when working in 96 to output to -0.50 ( as once suggested by Matt ) .. the l2 is only there for safety and doing nothing at all or just max. limit is 1 dB ... it's great to work / essential to me to hear the whole processing ... including limiting .. so compensating for that if needed with EQ .. maybe with these dance tracks it should have been max output -0.8 on the L2-hardware
because off the more extreme processing ...
In this case I just lowered the volumes on the processed 24/96 files with wavelab before src and going to 16

it's also great/ideal to have a high resolution mastered track then to get "stuck"with the lower rate ... making Mp3 or making revisions with digital gear only .. all can be done in the higher resolution .... just in case .. most off the times I prefer start from zero ..

the analyze function from wavelab also detects them ( 1 sample overs ) if you set the threshold to 1 instead off the default 4 samples ...

RME digicheck is great ... analyzers/metering/k12 and much more

thanks for sharing you're thoughts/expirience

[aivoryuk]

Originally Posted by MattGray

4. Remove the limiter & simply use your A/D to capture at 24-44.1 clipping on the way in instead of limiting. Once captured assemble your master do your fades etc & then lower the final output to -0.3dbfs prior to the final 16bit dithering. I bet in most cases with the right A/D this will sound better then the same gain reduction with your L2. You need a good A/D for this & obviously you can't push it too far (peaks only!). The benefits are this.. You avoid an SRC process & therefore the intersample peak issue. ...

Can Bob or anyone clarify if intersample peaks happen just in the SRC process???
In experiements where i had a 44.1 file and have used a limiter such as psp or L2 I still get intersample peaks. So I am under the impression that if even if you clipped the a/d at 44.1 as suggested above, you will still be subject to intersample peaks

[Spartacus]

Originally Posted by aivoryuk

Can Bob or anyone clarify if intersample peaks happen just in the SRC process???
In experiements where i had a 44.1 file and have used a limiter such as psp or L2 I still get intersample peaks. So I am under the impression that if even if you clipped the a/d at 44.1 as suggested above, you will still be subject to intersample peaks

as far as you get some consecutive 0dbFS samples, your DAC if it doesn't get enough headroom will create a signal over 0 because the peak analog signal reconstructed by the DAC will exceed this level.

so it's not only a SRC question, it's just a Converter issue.

[inlinenl]

Originally Posted by aivoryuk

Can Bob or anyone clarify if intersample peaks happen just in the SRC process???
In experiements where i had a 44.1 file and have used a limiter such as psp or L2 I still get intersample peaks. So I am under the impression that if even if you clipped the a/d at 44.1 as suggested above, you will still be subject to intersample peaks

Alex, I think you can clip the AD as much as you want but after that put the limiter in .... what was the ouput level off the limiter 0.00 dB or -0.30 dB ??? ... if the limiter is the final proccessing and the ouput is < 0.00 db there will not be any intersample peaks off > 0.00
IMHO

[Spartacus]

Originally Posted by inlinenl

Alex, I think you can clip the AD as much as you want but after that put the limiter in .... what was the ouput level off the limiter 0.00 dB or -0.30 dB ??? ... if the limiter is the final proccessing and the ouput is < 0.00 db there will not be any intersample peaks off > 0.00
IMHO

Even if the limiter doesn't see any information above 0dBFS in the digital world, still the DAC (especially cheap ones) may create >0 from consecutive maximum level samples in the reconstructed analog signal!

[aivoryuk]

Originally Posted by inlinenl

Alex, I think you can clip the AD as much as you want but after that put the limiter in .... what was the ouput level off the limiter 0.00 dB or -0.30 dB ??? ... if the limiter is the final proccessing and the ouput is < 0.00 db there will not be any intersample peaks off > 0.00
IMHO

even with the limiter set at 0.00 or at -0.3 there are still intersample peaks confirmed by metering on the psp xenon. thats ITB by the way so im using the input to boost the level. I haven't tried clipping the a/d as of yet due to a technical problem so am curious to know if by clipping a a/d at at 44.1 will produce intersample peaks. I think it will.

[inlinenl]

Originally Posted by Spartacus

Even if the limiter doesn't see any information above 0dBFS in the digital world, still the DAC (especially cheap ones) may create >0 from consecutive maximum level samples in the reconstructed analog signal!

Spartacus the limiter is digital .. after the AD ( clipped ) ... it's getting > 0.00 on it's input .. but limited to -0.30 on the digital output .. going to de DAC and to the input on the capturing DAW .. that dac/inputs is never going to see any 0.00 IMHO ....
please correct me if I'm wrong

[UnderTow]

Originally Posted by inlinenl

Spartacus the limiter is digital .. after the AD ( clipped ) ... it's getting > 0.00 on it's input .. but limited to -0.30 on the digital output .. going to de DAC and to the input on the capturing DAW .. that dac/inputs is never going to see any 0.00 IMHO ....
please correct me if I'm wrong

If the limiter doesn't look at inter sample peaks (like the L2), it will probably generate data with inter sample peaks at the output (and thus at the DAC) when processing clipped audio. If you set the output at 0.3, most of those inter sample peaks will probably be below 0 dB FS but a few will likely be above 0 dB FS.

Alistair

[PatrikT]

Originally Posted by UnderTow

If the limiter doesn't look at inter sample peaks (like the L2), it will probably generate data with inter sample peaks at the output (and thus at the DAC) when processing clipped audio. If you set the output at 0.3, most of those inter sample peaks will probably be below 0 dB FS but a few will likely be above 0 dB FS.

Strangely, when I limit at a work-pace of 44.1 kHz, with the output at -0.3 dB, there are no overs at all according to either the RME Digicheck or that funky little SSL-meter.

Even if the input to the limiter is 0.0 dB.

If I encode the wav to mp3 and reimport it to the DAW the lights will shine up like a epileptic volcano though...really scary. Makes one wonder what the hell people are paying for at the mp3-stores...


Best Regards
Patrik

[bob katz]

Originally Posted by aivoryuk

Can Bob or anyone clarify if intersample peaks happen just in the SRC process???
In experiements where i had a 44.1 file and have used a limiter such as psp or L2 I still get intersample peaks. So I am under the impression that if even if you clipped the a/d at 44.1 as suggested above, you will still be subject to intersample peaks

You are correct. Sample rate is irrelevant.

Any time you filter, you can get peaks that are HIGHER than the source, so it doesn't matter what the sample rate is, this is just a law of nature. It can be explained simply by saying that the higher harmonics may be out of phase with the fundamental and lower harmonics, so when you remove the higher harmonics, the level goes up!

[dynomike]

Originally Posted by PatrikT

Strangely, when I limit at a work-pace of 44.1 kHz, with the output at -0.3 dB, there are no overs at all according to either the RME Digicheck or that funky little SSL-meter.

Even if the input to the limiter is 0.0 dB.

If I encode the wav to mp3 and reimport it to the DAW the lights will shine up like a epileptic volcano though...really scary. Makes one wonder what the hell people are paying for at the mp3-stores...


Best Regards
Patrik

the mp3 format normalizes to 0 automatically. i don't know any way to get around that, so if i'm making mixes for the web, i'll just set the limiter output to 0. its going there anyway.

[UnderTow]

Originally Posted by dynomike

the mp3 format normalizes to 0 automatically. i don't know any way to get around that, so if i'm making mixes for the web, i'll just set the limiter output to 0. its going there anyway.

Some encoders might normalise but this isn't correct behaviour. I use CDex with LAME encoder and it certainly doesn't normalise.

Alistair

[PatrikT]

Originally Posted by UnderTow

Some encoders might normalise but this isn't correct behaviour. I use CDex with LAME encoder and it certainly doesn't normalise.

No, but it seems to go into clip/over mode any way.

Just like with the B/W-change from 96 to 44.1 kHz post limit.



Best Regards
Patrik

[aivoryuk]

Originally Posted by bob katz

You are correct. Sample rate is irrelevant.

Any time you filter, you can get peaks that are HIGHER than the source, so it doesn't matter what the sample rate is, this is just a law of nature. It can be explained simply by saying that the higher harmonics may be out of phase with the fundamental and lower harmonics, so when you remove the higher harmonics, the level goes up!

excellent, thanks for clearing that up for me Bob.

[MattGray]

Originally Posted by arf

Well, all modern ADCs digitize at oversampled rates and use on board real-time SRC to get down to the target rate. So it comes down to whether the SRC in the converter is preferable to SRC done externally with software or another digital device. That, of course, depends on what converters and/or software you have.

I guess I should have clarified by saying that if you are going out to analog processing then you need to capture through an A/D anyway so why not set the destination sample rate there on the A/D instead of doing a software SRC after conversion which will produce intersample peaks.

My experience with the HEDD's A/D is that it sounds better capturing at 44.1kHz then capturing at 96kHz & then using a software SRC later (if the destination is CD). The other advantage with the HEDD is that using a little of the HEDD's processing on the input side will lower the digital output by a few points. bringing it slightly under 0.0dbfs without the intersample peaks you usually find with using software SRC's. So you get a crew cut & a -0.3dbfs output straight to your DAW at 44.1kHz in real time & it sounds great.

I'm sure the Lavry Gold, Prism or Weiss would possibly do this even better. For these converters clipping would produce a 0.0dbfs capture so you'd have to either lower the level post conversion or use something like Ozone's 'intersample peak prevention', or PSP Xenon's oversample feature to reduce the chance of producing intersample peaks on cheaper consumer DAC's.

Originally Posted by arf

Good advice, but can't intersample peaks be on the order of a few dB or more?

So why is there so much talk about lowering .2dB - what good is that going to do for over-limited/over-clipped induced distortion in cheap a consumer DAC?

Depends on how flat topped the waveform peaks are, how high the RMS level is & how close to 0.0dbfs the peaks are as to how they will fair through a given D/A's reconstruction filter. Obviously using a simulated digital reconstruction filter in the form of an 'intersample peak meter' will give you a very good idea of how much you need to lower the peak level to compensate for these overs. There is other solutions like the Ozone & Xenon oversample features which tailor or reshape the edges of clipped peaks to prevent them from producing intersample peak overs. If limiting or clipping to high RMS levels then you may need to use this as well as dropping the peak level by a few tenths to prevent overs in cheaper DAC's. The key is to monitor these peaks so you know how much you need to lower the level or what approach to take to reduce it to an acceptable level.

[arf]

Originally Posted by MattGray

My experience with the HEDD's A/D is that it sounds better capturing at 44.1kHz then capturing at 96kHz & then using a software SRC later (if the destination is CD). -SNIP-

I'm sure the Lavry Gold, Prism or Weiss would possibly do this even better.

I'll confirm that capturing from analog at 44.1 using the Lavry Gold has worked very well even with some digital EQ and limiting afterwards. Call me crazy, but I find there are ways to get plenty of clean sounding level without clipping the ADC. I think clipping the ADC is a technique that grew out of legacy mastering where, at one point in time, the only new thing that was added to the chain was the ADC. It seemed to take legacy operations quite a few years to become facile with processing in the digital domain after capture.

[bob katz]

Originally Posted by dynomike

the mp3 format normalizes to 0 automatically.

I haven't noticed this. Is this really the case?

BK