Offline Upsampling

[audiventory]

There are two popular algorithms of resampling non-multiple sample rates: interpolation and over/down multiple resampling.

 

Interpolation

 

Interpolation is restoring of virtual inter-samples (between «real» samples).

 

Some resampling algorithms can use interpolation for non multiple resampling.

Like spline function, as example: Spline (mathematics) - Wikipedia, the free encyclopedia

Interpolation provide some non-linear distortions (artifacts).

Level of these artifacts depend on interpolation quality.

 

 

Over/down multiple resampling

 

For resampling popular audio sample rates also used algorithm without interpolation.

Input sample rate up to frequency that multiple for input and output sample rates.

After the frequency divide to output sample rate.

 

Example:

1. Upsampling 44 100 Hz x 160 = 7 056 000 Hz

2. Filtration 0 … 24 000 Hz passband (half of output sample rate 44.1 kHz)

3. Downsampling 7 056 000 Hz / 147 = 48 000 Hz

4. Filtration 0 … 22 050 Hz passband (half of input sample rate 44.1 kHz)

 

For conversion multiple sample frequencies we do:

1. Filtration 0 … 22 050 Hz passband (half of output sample rate 44.1 kHz)

2. Downsampling 88 200 Hz / 2 = 44 100 Hz

 

or

1. Upsampling 44 100 Hz x 4 = 176 400 Hz

2. Filtration 0 … 22 050 Hz passband (half of input sample rate 44.1 kHz)

 

Before upsampling we fill area between «real» samples by zeroes.

As interpolators used filters.

 

Here no direct sources of non-linear distortions. However non-linear distortions present at output of 4th stage.

Also conversion artefacts shifted/mirrored to audible range.

 

We control artifact's intensity via quality of filters. Here all filters are important.

 

 

 

Resume

 

1. For algorithms based on interpolation (like spline) generally being difference between multiple and non-multiple resampling.

 

Due appearing «virtual» samples, that approximately match to ideal (true) values in its point of time.

 

 

 

2. For algorithms based on multiple up-/down- sampling technically no difference between multiple / non-multiple resampling.

 

Due in both cases as interpolators used filters, that haven't non-linear distortions itself.

 

However quality of the filters impact to non-linear distortions due degree of artifacts suppressing.

 

Thus quality can distinguish for different combinations of input/output sample rates.

 

Degree of distinguishing don’t depend on multiple of not input and output sample rates.

 

Best regards,

Yuri Korzunov

[audiventory]

New AuI ConverteR 48x44 v.4.1.20 has improved stability of reading metadata (album, track In earliername, performer,...) from Internet CD database during CD ripping.

Read more More stability of reading CD database, ISO Conversion Control. New AuI ConverteR 48x44 v.4.1.20

[audiventory]

ANALYSIS: DSD-to-PCM 2015 - foobar SACD Plug-In, AuI ConverteR, noise & impulse response...

 

Comparision DSD to PCM converters

 

Archimago's Musings: ANALYSIS: DSD-to-PCM 2015 - foobar SACD Plug-In, AuI ConverteR, noise & impulse response...

[audiventory]

Applied 3rd generation filters for all input / output sample rates 44 kHz New AuI ConverteR 48x44 v.4.1.22. Applied 3G resampling filters for 44.1 kHz

 

[btlancaster24]

How does Sonic Studio "Process" and NCH "Switch" factor into this topic? Has anyone tried either of these programs and if so - how do they compare to the other software in this thread? I've tried both of them up to 24/96 AIFF and played back converted files on Amarra and Pure Music but having a hard time hearing a difference compared to 1411/44.1 AIFF, 256/44.1 AAC, and 192/44.1 MPEG.

[audiventory]

Hi btlancaster24,

 

Possibly you need try resample original file to maximal sample rate of your DAC. Bit depth better DAC native.

 

And should be supplied bit perfect playback in player and operation system - without resampling in drivers, player, anywhere.

 

Upsampling don't guarantee sound improving due many depend on system "software and hardware".

 

Best regards,

Yuri

[btlancaster24]

Thanks for your reply, Yuri. 24/96 is the max for my DAC. That's what I converted my original (1411/44.1 AIFF, 256/44.1 AAC, & 192/44.1) files to during the demos for SS "Process" & NCH "Switch". Then played them back at the new native rate in Amarra, Pure Music, JRiver, HQ player, Audirvana+ and Fidelia.

[audiventory]

As I assume not only resampling give sound advantages.

 

As example in my resampling software I use additional processing that allow perform "half of work" of DAC. Also qualitative offline conversion do that DAC do inside. However almost without time and resource limits. Sometime it give some sound improving.

 

For me, as example, better check sound on classical music - quiet places, jazz hi hats, acoustic guitar, trumpets.

 

Also 24/96 sweep sine 0...48 kHz allow check software hardware abilities. For right playback sine must sound pure up to 16 ... 18 kHz. After must be silence.

[btlancaster24]

Also 24/96 sweep sine 0...48 kHz allow check software hardware abilities. For right playback sine must sound pure up to 16 ... 18 kHz. After must be silence. ~Yuri

 

I don't get what you're referring to. Could you explain please?

[audiventory]

Yes.

 

Sweep sine is sine with frequency grow in time.

 

For digital signal (sample rate 96kHz) frequency the sine changed from 0 ... 48 kHz (96 kHz/2).

 

While the frequency placed in audible range (0 ... 20 kHz) usually no problem.

 

When the frequency came above 16 ... 20 kHz, we should not listen anything. It above our hearing limit. It's ultrasound.

 

However if software and/or apparatus has troubles with filtration or sample rate conversion, ultrasound mirrored/shifted to audible range. And we listen sweep sine (frequency can decrease instead growth) up to end of test file.

 

Same effect possible due nonlinear distortions of apparatus.

 

Some artefacts possibly listen after increasing volume. Here need be careful with ears. Possible unexpected sharp increasing of signal level.

 

Today I thought about taking video about it.

[btlancaster24]

I'm still trying to get a grasp on these things w/ computer audio, etc. and am not familiar with most of the terminology you're using and how to apply it. Thanks though for trying to help me. I appreciate it. All I know about upsampling is that there are programs that are supposed to make your music sound better by converting specs to your DAC's maximum bit depth / sample rate within players or offline software programs. And how to follow easy directions that don't confuse me.

[audiventory]

Hi Btlancaster24,

 

Here links to my little articles about oversampling

How Convert Sample Rate. Oversampling

 

and downsampling

How Convert Sample Rate. Divide. Downsampling

 

May be graphs and comments there help more clear illustrate my previous post.

 

Best regards,

Yuri

[btlancaster24]

thanks, Yuri!

[mitch751]

I am very interested with it and downloaded the software for a trial.

 

However the DSF file that I converted, the sound is pretty odd and something like out of phrase ie just listening in a hall with many echo ?

[audiventory]

I am very interested with it and downloaded the software for a trial.

 

However the DSF file that I converted, the sound is pretty odd and something like out of phrase ie just listening in a hall with many echo ?

 

Hi Mitch751,

 

1. What input sample rate you use?

2. What DSD sample rate?

3. What audio player software?

 

If possible send me log files straight after conversion to [email protected]

 

Best regards,

Yuri

[mitch751]

Input sample rate : 16/44

DSD : 128

Audio software : audiravana plus 1.5

[audiventory]

Input sample rate : 16/44

DSD : 128

Audio software : audiravana plus 1.5

 

Mitch,

 

I checked conversion FLAC 16/44 stereo to DSF 5.6 MHz. Playback is correct.

 

1. What CPU loading during playback? Recommended 3-15%.

 

2. What input file format?

 

3. What settings playback buffer?

 

4. Other DSD128 playback correctly?

 

Best regards,

Yuri

[audiventory]

Updated AuI ConverteR 48x44 v.4.2.

 

1. Applied 3G filters (3rd generation) for 96 kHz

 

2. Applied export text metadata to ALAC

 

3. Applied user defined temporary path in settings for distributing load between several disks

 

4. Added output AIF extension

 

5. Applied conversion priority for release resource for other applications or use all resources for conversion.

 

6. Applied Pause/Continue conversion for Audiophil-E edition and above.

 

7. Improved procedure downloading and installing of codecs

 

More info: New AuI ConverteR 48x44 v.4.2. Applied 3G resampling filters 96 kHz, pause conversion

[mitch751]

Oh dear. It turns out to be a problem of my DAC. Sorry about it. Thank you for all the help

[audiventory]

Oh dear. It turns out to be a problem of my DAC. Sorry about it. Thank you for all the help

 

Hi Mitch751,

 

Thank you for feedback. For me important know problem solving result.

 

Best regards,

Yuri

[audiventory]

New AuI ConverteR 48x44 v.4.2.1. Fixed launch crash for some conditions New AuI ConverteR 48x44 v.4.2.1. Fixed launch crash for some conditions

[audiventory]

New high resolution audio converter AuI ConverteR 48x44 v. 4.3. Applied all 3G (third generation) resampling filters. Improved stability of sigma delta modulator (DSF output files)

 

New AuI ConverteR 48x44 v.4.3. Applied 3G for all resampling filters

[Jud]

Thanks Yuri, look forward to trying it tonight. Does 3G for "all filters" include minimum phase?

[audiventory]

Hi Jud.

 

While linear phase only.

 

I'm glad last results. Now in average noise floor less 10 dB comparing 2G.

 

While I think: what do first:

 

Linear phase or full band option.

 

Currently I cut all above 20 kHz I.e. make work of digital filter of DAC in my software. I suppose it is one of reasons why AuI has good listening results. In combination with carefully tuned filter of course.

 

In 3G filters I use 20 ...24 kHz transient band for 48x kHz range.

 

I plan add intermediate option: cut all above 24...25 KHz.

 

If release full band will possibly compare my hypothesis about advantages of cutting above 20 kHz.

 

What you think about it all?

 

Best regards,

Yuri

[Jud]

I have a thought but no time - will write later tonight.