Downsampling hi-res to 44.1 with nothing in higher frequencies. Any loss?

[Killer3737]

This is going to sound silly, it sounds silly to me, but hey, if you don't ask you'll never know for sure.

So say I've got a stereo audio file that is native 48khz or higher. If I open that file in an audio editor and see there isn't actually anything above 22.05khz in the channels, and downsample it to 44.1khz, will the resulting file still be lower quality? By that I mean obviously I'm not actually trimming anything away, but will the frequencies that are present be affected negatively by the downsample?

I guess basically I'm asking if I should still archive the native 48khz version even if there is no benefit, or if a 44.1khz version will be audibly identical?

I said it was silly!

[smrpix]

You're conflating the digital temporal sampling frequency with the pitch tonal frequency. They're not the same thing.

[jman98]

You may be making a false assumption that the 44.1 kHz version will be approximately 10% smaller than the 48 kHz version. That's not the case. The file sizes will be just about identical. There's not really a compelling reason to downsample 48 kHz files to 44.1 kHz unless you have some sort of playback device you can't live without that can't handle 48 kHz audio. The only reason I bring this up is that people often assume that "lower sampling rate = much smaller file" and that's the case. You haven't given a compelling reason for doing the downsampling, so I'm forced to make assumptions about why you might want to do this. Apologies if that's not the case.

Will the 44.1 kHz be of lower quality? Well technically speaking it MIGHT be of very slightly lower quality due to the lower sampling rate, but I suspect that in a proper test you would very likely not be able to tell any difference. Some people might argue that the lower sampling rate would result in a very slightly inferior final result, but in practical terms there's not much difference. Since you're not going to save any file space with the lower sampling rate and the audio should be more or less the same, there's not a good reason for doing this.

[pandy]

This is going to sound silly, it sounds silly to me, but hey, if you don't ask you'll never know for sure.

So say I've got a stereo audio file that is native 48khz or higher. If I open that file in an audio editor and see there isn't actually anything above 22.05khz in the channels, and downsample it to 44.1khz, will the resulting file still be lower quality? By that I mean obviously I'm not actually trimming anything away, but will the frequencies that are present be affected negatively by the downsample?

I guess basically I'm asking if I should still archive the native 48khz version even if there is no benefit, or if a 44.1khz version will be audibly identical?

I said it was silly!

Question is OK, answer a bit difficult as it can be: No and Yes.

No - if your input signal have no content over 44.1k/2 then there is no signal loss.
Yes - samples are different (they have different values than in signal sampled with different sampling speed) and as such some people may consider that this is worse.

From mathematical perspective however there is no difference.

Higher than necessary sample rate provide more space for filter transition - it is difficult to create very sharp filters with high attenuation in stop band - such filters are usually digital as analog filter with desired parameters will be insanely expensive and long term probably unstable.

[hogger129]

The point of 48khz is to push filtering out of the audible range. But I would just leave it as is. Reducing the sampling rate to 44.1khz is not going to significantly impact the file size.

[pandy]

The point of 48khz is to push filtering out of the audible range. But I would just leave it as is. Reducing the sampling rate to 44.1khz is not going to significantly impact the file size.

There no significant difference in digital filter between 48 and 44.1. 44.1 is usually limited CD and 48 for other uses.

[-Habanero-]

This is going to sound silly, it sounds silly to me, but hey, if you don't ask you'll never know for sure.

So say I've got a stereo audio file that is native 48khz or higher. If I open that file in an audio editor and see there isn't actually anything above 22.05khz in the channels, and downsample it to 44.1khz, will the resulting file still be lower quality? By that I mean obviously I'm not actually trimming anything away, but will the frequencies that are present be affected negatively by the downsample?

I guess basically I'm asking if I should still archive the native 48khz version even if there is no benefit, or if a 44.1khz version will be audibly identical?

I said it was silly!

Because 48 is not perfectly divisible by 44.1, there will always be rounding imperfections in the downsample but extremely insignificant let alone audible. It's 2015 and most audio editors use very good anti-aliasing when changing samplerate so if you wanna downsample, you can.
I would be more worried about the dithering to be honest. After you downsample you will have to save the file which will add more noise if dithering is on and distortion if dithering is off. It won't be audible on most parts but will be on quiet parts if you amplify.
So if you have no good reason to mess with the file, don't.

@jman98
Don't answer questions he didn't ask. He specifically asked about a 48khz file with no content above 44.1khz and whether or not a downsample to 44.1khz will affect the frequencies below it.

[pandy]

Because 48 is not perfectly divisible by 44.1, there will always be rounding imperfections in the downsample but extremely insignificant let alone audible. It's 2015 and most audio editors use very good anti-aliasing when changing samplerate so if you wanna downsample, you can.
I would be more worried about the dithering to be honest. After you downsample you will have to save the file which will add more noise if dithering is on and distortion if dithering is off. It won't be audible on most parts but will be on quiet parts if you amplify.
So if you have no good reason to mess with the file, don't.

If you can multiply then you can also divide without rest only common denominator is high... this is base for all decent resampling algorithms - first signal is massively oversampled (samples are or just repeated or repeated and separated by zeros - ZOH vs FOH concept)and later massively decimated, as oversampling can be done mathematically lossless then decimation is usually lossy (accuracy limited) - however with modern algorithms decimation can be considered virtually lossless.

In real life classic example massively oversampled signal is for example SACD with DSD stream (in fact most of the modern ADC/DAC are single to maximum few bits systems and they provide multibitdepth trough oversampling/decimation) , to receive PCM samples signal is decimated in digital filter (and this is place where single bit DSD is turned to multiple bits PCM).

But once again 44.1 and 48 are comparable and main difference between systems are irrelevant from average ear perspective (fractionally lower noisefloor for 48kHz )

[newpball]

(fractionally lower noisefloor for 48kHz )

Could you explain that?

[pandy]

(fractionally lower noisefloor for 48kHz )

Could you explain that?

http://en.wikipedia.org/wiki/Oversampling
http://en.wikipedia.org/wiki/Oversampling#Noise

In this case we assume that useful audio signal lay up to 20kHz so anything higher than 2x20kHz can be considered as oversampling - as such 48kHz is marginally higher sample rate from 20kHz perspective and it will provide slightly lower noise floor and slightly better SNR but both are marginal as difference between 44.1kHz and 48kHz is very small - large OSR by definition improve SNR and lower noise flooor - massively OSR system like in case of SACD where 1 bit audio have 64*44.1kHz so SACD may have relatively low noise floor and provide relatively wide bandwidth (but only for high level signals -few dBFS).

[newpball]

In this case we assume that useful audio signal lay up to 20kHz so anything higher than 2x20kHz can be considered as oversampling - as such 48kHz is marginally higher sample rate from 20kHz perspective and it will provide slightly lower noise floor and slightly better SNR but both are marginal as difference between 44.1kHz and 48kHz is very small - large OSR by definition improve SNR and lower noise flooor - massively OSR system like in case of SACD where 1 bit audio have 64*44.1kHz so SACD may have relatively low noise floor and provide relatively wide bandwidth (but only for high level signals -few dBFS).

So then there is noise and there is oversampling and those are not considered the same?

So if a signal does not contain anything over 20kHz the rest is divided in an oversampling and noise part?
Is that basically what you are saying?

[pandy]

So if a signal does not contain anything over 20kHz the rest is divided in an oversampling and noise part?
Is that basically what you are saying?

Nope - sampling as process introduce noise (distortions) http://en.wikipedia.org/wiki/Quantization_%28signal_processing%29#Quantization_noise_model - oversampling reduce this noise.

[newpball]

So if a signal does not contain anything over 20kHz the rest is divided in an oversampling and noise part?
Is that basically what you are saying?

Nope - sampling as process introduce noise (distortions) http://en.wikipedia.org/wiki/Quantization_%28signal_processing%29#Quantization_noise_model - oversampling reduce this noise.

Can't a bandwidth-limited signal, in this case 0..20kHz, be perfectly reconstructed if sampled above the Nyquist rate?

[pandy]

Can't a bandwidth-limited signal, in this case 0..20kHz, be perfectly reconstructed if sampled above the Nyquist rate?

Not sure what are you asking for?
By definition any bandwidth limited signal can be reconstructed if sampling rate is at least 2 times higher than bandwidth (so this apply also to bandpass sampling).

[newpball]

Can't a bandwidth-limited signal, in this case 0..20kHz, be perfectly reconstructed if sampled above the Nyquist rate?

Not sure what are you asking for?

You are claiming that a signal from 0..20kHz sampled at 44 has more noise than one sampled at 48. If the signal already can be perfectly reconstructed at 44kHz what in your opinion is the extra 4kHz going to do?

[pandy]

Can't a bandwidth-limited signal, in this case 0..20kHz, be perfectly reconstructed if sampled above the Nyquist rate?

Not sure what are you asking for?

You are claiming that a signal from 0..20kHz sampled at 44 has more noise than one sampled at 48. If the signal already can be perfectly reconstructed at 44kHz what in your opinion is the extra 4kHz going to do?

This is more problem of rendering device (DAC) - some systems prefer to have one frequency more than others, different systems don't care - classic example is EMU10k chip where native sample rate was 48000 and this was recommended.

48 seem to be more PC friendly, 44.1 seem to be more CD friendly. If system can accommodate 48 then 48 will be superior to 44.1 (especially when source is 48 then no resampling is involved).

[newpball]

48 seem to be more PC friendly, 44.1 seem to be more CD friendly. If system can accommodate 48 then 48 will be superior to 44.1 (especially when source is 48 then no resampling is involved).

True.

Of course the industry long ago moved away from recording audio 44/48 and 16 bits. 96k / 24bit is the minimum now.

Consumer audio is more or less forced to still use those archaic values.

With respect to noise, I think you are actually right 48 will have a lower nose level than 44 because the digital noise is spread out over the whole range of the signal, a higher sampling rate will impact a smaller percentage of the given bandwidth, 48 over 44 would calculate to about a 0.4db better noise level. Far better is it to increase the bitdepth to reduce noise.

[hello_hello]

In line with what hogger129 said, as far as I'm aware the main reason for a 48k sampling rate compared to 44.1k is because in order to sample a signal accurately it needs to be sampled at double the highest frequency. If you want to sample up to 20kHz, you need to sample at 40k, so at 44.1k and you need a low pass filter that starts rolling off at 20kHz and lets nothing get past it over 22kHz. Sample at 48k and you need a filter that starts rolling of at 20kHz and lets nothing get past it higher than 24kHz. From what I understand, the second filter is easier to build.

You are claiming that a signal from 0..20kHz sampled at 44 has more noise than one sampled at 48. If the signal already can be perfectly reconstructed at 44kHz what in your opinion is the extra 4kHz going to do?

Of course the industry long ago moved away from recording audio 44/48 and 16 bits. 96k / 24bit is the minimum now.
Consumer audio is more or less forced to still use those archaic values.

If the signal can already be perfectly reconstructed at 44kHz what in your opinion is the extra 48kHz going to do?

Of course when mixing and adding effects etc, a higher sampling rate/bitdepth is preferable to reduce rounding errors, but calling 44/48k and 16 bits archaic for consumer audio seems fairly odd to me when it appears to have been fairly well proven higher sampling rates and greater bit depths don't provide an audible difference.

Audibility of a CD-Standard A/D/A Loop Inserted into high-Resolution Audio Playback (pdf)

I found this one a little while ago, but haven't got around to reading it all yet.
24-Bit vs. 16-Bit Audio Test


Killer3737,
You might find these videos interesting.
http://xiph.org/video/

[SyncroScales]

I have a question. I have taken 48kHz audio, did some editing on it and exported with 44.1kHz.

The 44.1kHz was distorted and "sounded like a wall of noise".

It was like I had not equalized each track of a song in the mixer to remove frequencies that were not being used by the sounds.

Eg: Kick drum was in 60, 120, 320 Hz range. I should lower 16K down to 800 Hz to zero to isolate the sound of 60, 120 and 320 Hz.
Or Keys and hi hats were in 3K, 5K, 12K and 16K range, so I should equalize keys and hi hat with 1K - 60 Hz at zero.

This way the audio does not "fight for sound space"?

Am I talking about something else?

Note: When using 44.1kHz wav, mp3, etc source material/sound and exporting as 44.1kHz audio with the same editing, the distortion does not happen.

[pandy]

I have a question. I have taken 48kHz audio, did some editing on it and exported with 44.1kHz.

The 44.1kHz was distorted and "sounded like a wall of noise".

Use good sample rate converter (SRC), there is many various algorithms and some of them are relatively low quality.

@sox --multi-threaded --buffer 131072 -S -V -D %1 -b 16 %1_src2CD.wav rate -v -s -I 44100.0 gain -n -3.0102999566 dither -f shibata -p 16 stats

above good example for SRC with SoX ready to produce wave file compatible with CD Audio.

[SyncroScales]

What is SoX?

So can I still do edits, or should I just use a Sample Rate Converter with SoX (or something else) and then edit the Wav file?

At the point I have found not doing any edits do a 48 kHz file gives a good result when downsampling into a 44.1kHz file. Then editing.

[pandy]

What is SoX?

So can I still do edits, or should I just use a Sample Rate Converter with SoX (or something else) and then edit the Wav file?

At the point I have found not doing any edits do a 48 kHz file gives a good result when downsampling into a 44.1kHz file. Then editing.

I would recommend to use provided example as the end - there noiseshaping there and requantization to 16 bits.
Usual way is perform all processing with higher sample rate and bitdepth than target.

[hello_hello]

There's a SOX re-sampling DSP you can use with foobar2000's converter (foobar2000 comes with one called Resampler PPHS but apparently SOX is better). No reason why you couldn't convert the audio to a wave file while resampling it (you can add DSPs to the conversion path) and then edit it with whatever audio editer you prefer. Or edit it first, then convert/re-sample it.

You'd probably just want the "normal" version. http://www.hydrogenaud.io/forums/index.php?showtopic=67373



You can install DSPs via the File/Preferences menu under the "Components" section.