• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

How to avoid digital artefacts in recordings ?

Is the SRC done in DAW:s audible ?

  • 1. No

    Votes: 3 75.0%
  • 2. Yes

    Votes: 1 25.0%
  • 3. Sometimes

    Votes: 0 0.0%

  • Total voters
    4

Tangband

Major Contributor
Joined
Sep 3, 2019
Messages
2,994
Likes
2,789
Location
Sweden
In another thread, there was some discussion about that the sample frequency should never change in a recording .

“One” should never do that. Everything should be recorded at the maximum available resolution and never changed. There’s no reason to do that. The only time it may be changed is at mastering as per the medium the track is distributed as.
In a purist perspective - you are absolutely right. But if you record something different than only classical instruments with two channels without effects , you have to mix it in a software program ( DAW ) like Logic , using reverb plug ins ( often 48 KHz ) or compression ( 48 KHz ) and so on….
What you hear in a ” normal ” production is something thats in the beginning maybe was recorded at 96 KHz but has been thru SRC many times before if finally arrives at TIDAL or Spotify at 44,1 KHz .

One must remember that putting some reverb and compression on a drumkit ( always done ) in a mix demands resampling 2 times for the whole track ( or often 8 tracks for only the drumkit ) .

There is a lot of confusion about this .

Me , - I always try to record acoustical instruments at native sampling rate at 96 KHz , using only two microphones, and skipping the whole mixing process.

But - I still have to do one digital manipulation before the recording is finished and thats ” normalisation” of the recording .

In an acoustical recording , you always have about -10 dB as a margin for digital clipping . The recorded tracks will be about -10 dB as loud as a normal CD . At normalisation, you lift up the level to -1 dB . This is done in digital domain with 32 or 64 bit resolution in a DAW.

My experience with DAWs like Logic Pro and Audacity is that this simple digital manipulation ( in Logic Pro X its done at 64 bit internaly, 32 bit in audacity ) can be heard as a little less natural sound , unfortunately.

The -10 dB 96 KHz 24 bit recording will sound a bit better than the finnished -1 dB recording . The -10 dB recording will sound more natural and with less digital ” glare” .

This is a sad thing - because when I started recording , I always thought this was entirely inaudible wich is not true.

————————

What do you, hobby-sound engineers or professionals working in studios have to say about this ?
 
Last edited:
OP
Tangband

Tangband

Major Contributor
Joined
Sep 3, 2019
Messages
2,994
Likes
2,789
Location
Sweden
This is a very good page :

Here you can see a comparison of two DAWS : Audacity and an earlier Reaper doing 96 to 44,1 kHz SRC. There is big differences.
AC5B763A-C58C-48D8-A7BD-B50FB7D37701.png

If I use Audacity - I can still hear a difference where the converted material at 44,1 kHz sounds worse than the original at 96 kHz . It dont show up in measurements.
 
Last edited:

BeerBear

Active Member
Joined
Mar 9, 2020
Messages
264
Likes
252
SRC, being a lossy process, can do some harm. But...
There are plenty of tools and processes in production that are even more "damaging" than SRC:
- room noise
- mics and mic techniques
- amps and preamps
- AD and DA converters
- external (hardware) effects
- plugins
- MQA
- ....

Typically, all of those harm the fidelity of a recording more than SRC, often orders of magnitude more. So, I'd worry about those instead.

This is a very good page :

You can here see a comparison of two DAWS : Audacity and an earlier Reaper doing 96 to 44,1 kHz SRC. There is big differences.
View attachment 209581
Indeed, SRC quality can vary. But even some not-so-good performers might not have any audible effects in the end.

I recently posted an example of Windows' SRC, which is technically worse than good offline resamplers (but it's not terrible, either).
See if you can ABX those two files in a blind test. The file went through two Windows SRCs, and I still doubt that you can hear a difference.
 

sergeauckland

Major Contributor
Forum Donor
Joined
Mar 16, 2016
Messages
3,440
Likes
9,100
Location
Suffolk UK
I find it interesting that many digital audio mixers have an SRC at every input, to avoid the need for sources to be synchronised. This is especially important with remote sources like satellite feeds or Audio Over IP. Even if the incoming sample rate is nominally the same as the mixer's internal sample rate, the SRC is still used to maintain synchronism over time, as all clocks vary slightly, and timing errors build up over minutes or hours. There's never been any suggestion I've ever heard that current SRCs are anything but transparent, so I don't see it as a problem.

S.
 

DVDdoug

Major Contributor
Joined
May 27, 2021
Messages
2,920
Likes
3,834
I agree with BeerBear.... You are very unlikely to hear a difference in a proper, scientific, level-matched blind ABX Test. The differences will be very small and under normal conditions you won't even have A & B to compare so it doesn't matter. It's way-down at the bottom of the list of things to worry about. And, if you're resampling you probably have a good reason and it may be unavoidable.

Plus, audio production normally includes a lot of processing to intentionally change/improve the sound. As a listener you might want "bit perfect" audio reproduction but that doesn't apply to production.

In an acoustical recording , you always have about -10 dB as a margin for digital clipping .
Pros often leave more headroom but nothing bad happens if you get close to 0dB. Headroom is a funny thing... If you don't use it you didn't need, it and if you use it it's no longer headroom. And mixing is done by summation* so if you are multi-track recording the levels will usually have to be reduced sooner-or-later anyway.

You actually have less resolution at lower levels (because you aren't using all of the bits of your analog-to-digital converter). But, with digital recording (especially at 24-bits) you have far-more resolution than you need and digital recording levels are not that critical unless you clip or unless they are VERY low.

IMO - We get a little over-obsessed with "meters". In the analog days you wanted a hot signal to overcome tape noise. And analog tape soft-clips as you go over 0dB so it was not unusual to go occasionally into-the-red. With digital, we just need to avoid clipping. Analog-to-digital converters are absolutely limited to 0dB and they will hard-clip if you "try" to go over.

The recorded tracks will be about -10 as loud as a normal CD.
"Loudness" is complicated. Loudness doesn't correlate well with peak levels. It's more related to the short-term average. If you normalize all of your recordings some will still be louder than others. Most commercial recordings are dynamically compressed & limited (with make-up gain applied) to make them louder. Even without compression a normalized "dense" recording (like an orchestra or rock band) will sound louder than a piano or acoustic guitar.

At normalisation, you lift up the level to -1 dB .
A lot of CDs are normalized to 0dB. Some popular CDs are even slightly-clipped (to "win" the loudness war). And if you make an MP3, some peaks and other's decrease so some people leave 1 or 2dB of extra headroom for that. (MP3 can go over 0dB without clipping but you'll clip your DAC if you play it a "full digital volume".)

The -10 dB 96 KHz 24 bit recording will sound a bit better than the finnished -1 dB recording . The -10 dB recording will sound more natural and with less digital ” glare” .
That's just a perception. Of course it will sound different at different volume levels. There are only very tiny rounding errors with digital amplification. When you attenuate digitally (in an integer format) you loose resolution but that's not audible unless you make a HUGE attenuation, and it's still not audible unless you re-amplify.

A proper ABX test would require you to match the levels with an analog adjustment. And you also loose resolution when you attenuate in analog (because of a reduced signal-to-noise ratio) and if you amplify in analog you add noise and distortion. Within reason, none of this is normally audible and digital processing is almost always better.


* Mixing is actually more of a weighted-average since you have level controls for each channel, plus a master level control. But one "interesting thing" is, since you are summing you increase the "bits' so you are increasing digital resolution (although you usually attenuate and reduce resolution again). But, you can mix several 16-bit tracks and end-up with 24-bits of true-resolution in your final mix!

BTW - Analog mixers are built-around summing amplifiers.
 
Last edited:
OP
Tangband

Tangband

Major Contributor
Joined
Sep 3, 2019
Messages
2,994
Likes
2,789
Location
Sweden
I find it interesting that many digital audio mixers have an SRC at every input, to avoid the need for sources to be synchronised. This is especially important with remote sources like satellite feeds or Audio Over IP. Even if the incoming sample rate is nominally the same as the mixer's internal sample rate, the SRC is still used to maintain synchronism over time, as all clocks vary slightly, and timing errors build up over minutes or hours. There's never been any suggestion I've ever heard that current SRCs are anything but transparent, so I don't see it as a problem.

S.
Interesting .
There is some data to read from different SRC chips thats used , and many modern good chips have a good SINAD about -130 dB , so it should be audible transparent. A less good chip is inside the Yamaha wxc50 ( DSP YSS952) it has a SINAD about 108 dB according to Yamaha.

The external mixer digital hardware has also digital volume controls that controls every input and output. If its set to -36 dB as an example, you have lost 6 bits. So that -130 SRC data might be needed after all.
 
Last edited:

Waxx

Major Contributor
Joined
Dec 12, 2021
Messages
1,933
Likes
7,688
Location
Wodecq, Hainaut, Belgium
This is one of the big issues with PCM sampling (the technical proces to digitise audio) and processing and it's the reason why many prefer analog from start to finish, and why dsp is always a degrading of the sound quality. But the SRC agloritmes became that good that with most modern devices or software in reality it does not matter that much anymore in the bigger picture. And any transfer from media format causes little distortions, also between analog media. So just use quality systems and you should not worry to much about it.
 

LtMandella

Member
Joined
Jan 13, 2021
Messages
67
Likes
44
Location
Las Vegas
In another thread, there was some discussion about that the sample frequency should never change in a recording .


In a purist perspective - you are absolutely right. But if you record something different than only classical instruments with two channels without effects , you have to mix it in a software program ( DAW ) like Logic , using reverb plug ins ( often 48 KHz ) or compression ( 48 KHz ) and so on….
What you hear in a ” normal ” production is something thats in the beginning maybe was recorded at 96 KHz but has been thru SRC many times before if finally arrives at TIDAL or Spotify at 44,1 KHz .

One must remember that putting some reverb and compression on a drumkit ( always done ) in a mix demands resampling 2 times for the whole track ( or often 8 tracks for only the drumkit ) .

There is a lot of confusion about this .

Me , - I always try to record acoustical instruments at native sampling rate at 96 KHz , using only two microphones, and skipping the whole mixing process.

But - I still have to do one digital manipulation before the recording is finished and thats ” normalisation” of the recording .

In an acoustical recording , you always have about -10 dB as a margin for digital clipping . The recorded tracks will be about -10 dB as loud as a normal CD . At normalisation, you lift up the level to -1 dB . This is done in digital domain with 32 or 64 bit resolution in a DAW.

My experience with DAWs like Logic Pro and Audacity is that this simple digital manipulation ( in Logic Pro X its done at 64 bit internaly, 32 bit in audacity ) can be heard as a little less natural sound , unfortunately.

The -10 dB 96 KHz 24 bit recording will sound a bit better than the finnished -1 dB recording . The -10 dB recording will sound more natural and with less digital ” glare” .

This is a sad thing - because when I started recording , I always thought this was entirely inaudible wich is not true.

————————

What do you, hobby-sound engineers or professionals working in studios have to say about this ?
And then there are the famous RCA Living Presence recordings done in the 60's. Well known for their great sound, and DAW was not even imagined.
 
Top Bottom