Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Vincent Kars]

configuring Media Center for the best audio quality

That are sentences I love to read.
I expect a full 64 bit audio path in the near future!

WASAPI indeed yields a better sound quality at the expense of comfort.
Recently I had a track that failed to play in WASAPI, it was in mono.

I don't like programs to do all kind of under water activities.
I like to know what is going on.
A dialog like this appeals to me


It allows you to test and to exclude certain options.
In case of USB-audio this can be refined as the enumeration allows the app to fill in all possible options.
In case of SPDIF this is not possible.

I think MC should play all audio at its native sample rate.
If this fails, I prefer a dialog.
It should tell me the reason why it can't be played
It should offer the options
-to convert
-to stop
-to remember my choice
This solves the 'control' issue

Maybe a profile might help e.g.
Benchmark DAC 1 - SPDIF
Benchmark DAC 1 - USB
Allowing you to switch rapidly between configurations and/or audio endpoints.

Documentation.
I'm afraid indeed people don't read the Wiki and keeping it up to date is probably not the highest priority at JR.
In MC15 there are couple of dialogs already presenting a help text/explanation as a integral part of the dialog.
You might consider to expand this.
I prefer a collapsible help text as the dialogs become sometimes pretty big.

[Alex B]

I expect a full 64 bit audio path in the near future!

That is a good one. You would then be able to have up to a 384 dB dynamic range.

http://blogs.msdn.com/b/audiofool/archive/2006/08/22/713498.aspx   

[JimH]

We once tested what people could hear, with files at 8, 9, 10 bit depth, etc.  Very few people could hear the difference beyond 11 or 12 bits.

[Vincent Kars]

We once tested what people could hear, with files at 8, 9, 10 bit depth, etc.  Very few people could hear the difference beyond 11 or 12 bits.

If this is the case, do you expect them to hear the difference between DS and WASAPI?

64 bits
This is not about the audio file, 24 is max and 24 bits is 24*6=-144 dB FS
This is way below the noise floor of any gear (+/- 110 dB FS).

64 is about DSP
All top professional audio processing software use 64 bits.
Simply to keep the unavoidable quantization error way below the noise floor.
Now in a production environment you do far more calculations as when doing playback but e.g. sample rate conversion is often used in playback

This is a nice one: http://src.infinitewave.ca/
Compare e.g. the SRC of Ableton Live with iZotope 64 bits.

A comparison of SRC in OSX and Vista: http://www.thewelltemperedcomputer.com/Lib/OperatingSystemsHandlingOfSampleRates.pdf

A comparison of native mode and re-sampling in OSX by Benchmark Media: http://extra.benchmarkmedia.com/wiki/index.php/Sample-rate

I do think that there are technical advantages to have a 64 bit audio path and if you want to sell MC to the audiophile community, it is an asset.
There are  (not only technical) reasons why Ammara or Pure Music stress to have a 64 bit data path.

[JimH]

Vincent,
Please copy and paste the parts of your first post that were on the original topic.

[Matt]

It's easy to show, with math, that 32-bits provides enough headroom for a lot of DSP at 24-bit output before precision becomes any kind of issue.

With that said, we may change the Media Center playback engine to 64-bit, simply because there's probably not much downside.  

In many cases, computers are faster working with 64-bit decimals than 32-bit decimals.  Memory usage for buffers would double, but I don't consider that a big deal.

Due to how the playback pipeline is designed, it would be difficult to allow it to be user selectable between 32-bit and 64-bit.  I think we would just have to make the switch.  Version 17?

[Vincent Kars]

I think we would just have to make the switch.  Version 17?

You have 1 who is going to upgrade...

[Alex B]

From the other thread:

... Regarding the bit depth, it would be good to add an option for high quality dithering with noise shaping for the 16-bit mode.

24-bit is a no-brainer. It should be used if the device supports it and it does not need dithering (the quantization errors are very small and unlikely to be audible in any circumstances).

I think dithered and noise shaped 16-bit output would be more likely to audibly improve quality than 64-bit DSP when the audio device driver can't handle 24-bit input.

I have such a device on my new Asus mobo (VIA HD audio VT1708S). It is otherwise a decent sound device, but probably because of bad driver design 24-bit does not work when the analog outputs are used. I was able to tweak the Windows registry and get 44.1 kHz to work natively, but not 24-bit (the VIA chip itself supports both).

The internal volume, replay gain, EQ, bass management, room correction & other DSP features can eat a considerable amount of the 16-bit dynamic range that is available for the output. The quantization errors are more likely to become audible when MC is used as a preprocessor and volume level controller.

To demonstrate how dithering and noise shaping can improve audio quality I created two audio samples that contain the effect in an exaggerated form. First I selected two very quiet tracks and cut a short sample from each. Then I combined the samples and created two output files with 8-bit content by using MC's disk writer and the iZotope Ozone VST plugin. One file is simply truncated to 8-bit and the other one is dithered with iZotope Ozone using the audibly best noise shaped dithering option.

Naturally both files are noisy, but in the dithered file the noise is less vigorous and it does not contain the ugly distortions and dropouts that exist in the truncated file.

The files in the sample package:
8-bit_dithered_and_noise_shaped.ape
8-bit_truncated.ape
16-bit_unprocessed.ape (for comparison, directly outputted through MC's disk writer)

All APE files are in the 16-bit/44.1 KHz format. The 8-bit versions just do not contain anything in the lower 8 bits.

The sample package:
https://rapidshare.com/files/1987681784/dithered_vs_truncated_samples.zip (11.4 MB)

[Vincent Kars]

Matt

What I had in mind are articles like these: http://www.cadenzarecording.com/images/floatingdither.pdf

Now you probably have looked further into this matter than I ever will (and able to understand) but it gives me the feeling that precision will improve when moving to 64.

What do I understand wrong?

Just a random thought: a 64 MC audio engine running on a 64 bit Win with 16 Gb memory allowing you to load an entire Der Ring des Nibelungen, upsampled to 24/176 in memory.
Any audiophile player who could match this?

[Alex B]

To Matt,

Apparently Monkey's Audio has a problem with compressing the sample files that have 8-bit content in a 16-bit PCM file. For instance, FLAC can compress them a lot more. See the attached screenshot.

[Matt]

From the other thread:
I think dithered and noise shaped 16-bit output would be more likely to audibly improve quality than 64-bit DSP when the audio device driver can't handle 24-bit input.

I think this is a separate issue from a 64-bit data path.

Dithering may be a good thing for 16-bit output if processing is done, although I would argue that it will be impossible to hear shaping of the 16th bit on any hardware that only supports 16 bits.  This really only applies to low-end hardware, since good hardware supports more than 16 bit output.

Using an 8-bit example shows the phenomenon, but also exaggerates it by 48 dB! compared to a 16-bit file.

[Alex B]

Using an 8-bit example shows the phenomenon, but also exaggerates it by 48 dB! compared to a 16-bit file.

You must remember that I don't have 16-bits anymore when MC's volume slider is at about the middle of the scale and I have volume leveling, surround processing, room correction and bass management enabled. I might have only 12-bits left.

How about this: Listen to my samples as loud as you can without feeling really unpleasant or causing ear damage and then reduce the volume level by about 24 dB (*) (4 more lost bits in addition to what may have been lost in the DSP processing). Can you still hear the difference?

With 16-bit audio a real life situation could be to listen to something like a classical symphony at a natural volume level. Then the peaks may be 100 dB SPL and the quietest passages 40 dB SPL. When heavy DSP processing is used the remaining dynamic range during the quietest passages may be small enough to make the distortion that is caused by truncation audible. Noise shaped dithering can increase the perceivable dynamic range considerably.

EDIT

(*) Since the peak volume in my sample is about -11 dB, it would be more correct to reduce the volume by -35 dB, but I think the difference would still be audible if the used amp has enough reserve to first achieve the "loud" volume level.

[Alex B]

I think this is a separate issue from a 64-bit data path.

I am also trying to bring this bit depth discussion down to earth. As you said, the math proves that the additional precision of 64-bit processing cannot make the audible quality better.

In any case we are splitting hairs when we are trying to improve things that are already quite good, but IMHO what I am suggesting can improve audio quality in a practical listening situation. Noise shaped dithering to 16-bit after DSP processing is a commonly accepted method to improve quality. It is a standard procedure in the audio industry.

Many users have these new "HD" audio devices on their motherboards. Generally they seem to perform fine, except that the 24-bit output may not work as promised.

[Matt]

How would it work if, in DSP Studio > Output Format > Bitdepth, you could select "16-bit" or "16-bit with dithering"?

Do you have a dithering algorithm of choice?  I know there's a lot of debate about the best shape, color, etc. of dither.

Thanks.

[HiFiTubes]

If you force native sample rates, folks with non-auto switching cards get stuck doing a manually change of their sample rate. Upsampling may not be beneficial but it can be useful in terms of usability; you could have crossfades etc. enabled and play back various sample rates uninterrupted.


What happens when the DSP is set to 32 bit but when using WASAPI for example you have to use 24bit (with the Lynx cards).

If you are upsampling, using room correction, Replay Gain, and Clip Protection in MC, I guess "technically" one should dither as the word length is changed from 32bit to 24 for the driver? the Lynx mixer has dither built into to each output strip actually.


In recording 64bit has some commonly discussed advantages, just not sure any can be found with regard to fidelity in the playback realm.

Antelope uses some 64bit technology in their DACs:

64-BIT TECHNOLOGY

Sophisticated 64-bit algorithms allow precise control over our Acoustically Focused Clocking (AFC) and Jitter Management; two critical components to improve audio via clocking. The incoming digital signal is brought to life with this unique 64-bit technology, resulting in smooth and detailed audio whatever the method of digital connectivity.

[newsposter]

and now we should start talking about bit depth at what effective clock rate........

[Vincent Kars]

How would it work if, in DSP Studio > Output Format > Bitdepth, you could select "16-bit" or "16-bit with dithering"?
Do you have a dithering algorithm of choice?  I know there's a lot of debate about the best shape, color, etc. of dither.

Some modern DAC’s offer user selectable filtering.
(simply, they use the same Wolfson chipset and one of the options is a couple of different filters)
I suggest to do the same with MC, a couple of different filters so the user can choose.
Option probably not limited to 16 but to all DSP where applicable

[HiFiTubes]

Well Matt has suggested that even for 16/44 playback to use 24bit in the DSP, so I'm confused about why we need to worry about dither given how little the DSP is going to affect anything, and it has plenty of headroom being that it is 32bit internal.

If I have native 24/96 material for example, I'm not going to dither because my driver and DAC and driver utilize the 24bit pipeline.

Confused.

[Matt]

In a coming build:
NEW: Added ability to dither output when outputting 8-bit or 16-bit data (selectable in DSP Studio > Output Format > Bitdepth).

We're using a TPDF dither.  It sounds noticeably better at 8-bit.

It is theoretically better at 16-bit if you are performing processing, but I can't hear it using pretty high end stuff with volumes that would destroy the equipment if a full amplitude signal hit it (*).

At 24-bit (or more) output, I think dithering is pointless.  We could offer it, but only instrumented testing of a digital signal could ever show it was there.

It is possible that shaping the noise to model the human ear would outperform a TPDF dither, although there's some disagreement about this.  In any event, I think we're back to Alex's comment about trying to perfect something that's subtle and already quite good.


* This test is not on my "top ten fun things to do list."  Maybe we need a summer intern to do the listening under headphones set so high they'd melt your face if a full strength signal accidentally came through

[newsposter]

* This test is not on my "top ten fun things to do list."  Maybe we need a summer intern to do the listening under headphones set so high they'd melt your face if a full strength signal accidentally came through

Or...... establish a standard test suite of quality monitors, amps, and microphones to simulate a disposable summer intern.  Feed MC output to the amps and capture what the mics hear, digitize it, do a comparison of what is fed to the amps/speakers.

This would give you some sort of repeatable basis for judging the effect on analog output (the speakers) of what you're doing on the digital (MC internals) side as well as the effectiveness of varying sound cards and component connect cabling.

[Matt]

Build 101 is out, and features a purely 64-bit data path for all audio.

This includes all DSPs that ship with the program and also VST plugins (assuming they support 64-bit).

More here:
http://yabb.jriver.com/interact/index.php?topic=59364.0