[Seidhepriest]

Must be an SSRC plugin option... Have you tried the Winamp setup?

 

[b0dhi]

In Winamp I use an ASIO output plugin that has SSRC built into it. It works fine there, but ofcourse...no linear mode. The winamp version doesn't have a hardware mixing option either, though.

 

[Seidhepriest]

Whee. So well... Does the F2000 SSRC do linear resampling?

The current versions of SSRC lacking an option for software or hardware resampling might mean it automatically switches if hardware resampling fails, or just does everything in software (it is safer, and hey, CPUs nowadays are powerful...).

According to Wiki, http://en.wikipedia.org/wiki/Free_Lossless_Audio_Codec , FLAC supports non-floating-point bit depth of up to 32-bit and any sampling rate up to 1048570 Hz. So as long as the wave format is integer-based, an external editor should be capable of linear resampling, then FLAC could re-encode the 192-KHz waves.

 

[b0dhi]

You're talking about SSRC and not SRC, right?

If so, no, it doesn't.

 

[Seidhepriest]

So you're pretty much stuck with SRC?

 

[Seidhepriest]

The way it looks...

1. A stack of Winamp DSP adapter for F2000+Adaptx 3.5 DX adapter+some kind of Directx linear resampler.

2. The same stack, except running natively in Winamp. DSP processing is done before output. Hence whatever the DSP resampler's going to pass to the output plugin, won't be affected by the plugin's resampling.

3. A linear resample in an external wave editor, then re-encoding to FLAC.

4. The SRC plugin author might reply?

 

[b0dhi]

Thanks for all your help. I'm going for "5. Write own plugin"

 

[Hancoque]

Quote:

Originally Posted by b0dhi /img/forum/go_quote.gif Hancoque, what artifacts are you talking about, and what upsampling rate did you use?

I created 32 bit float samples at 44.1 kHz and upsampled these to 192 kHz using the different plugins and also Adobe Audition. All calculations took place in 32 bit floating point precision (including foobar2000).

These are the results:
Spectral view (original)
Spectral view (upsampled by Adobe Audition)
Spectral view (upsampled by SRC linear)

Wave view (original)
Wave view (upsampled by Adobe Audition)
Wave view (upsampled by SRC linear)

 

[b0dhi]

The wave view for the original and upsampled waves is not accurate. The software you're using is using interpolation between samples when it's displaying the waveform, automatically. The wave view for "original" should look very much like the one for SRC linear.

The spectral analysis seems accurate though. As expected, the linear upsampling mode measures poorly.

 

[Hancoque]

Quote:

Originally Posted by b0dhi /img/forum/go_quote.gif The wave view for the original and upsampled waves is not accurate. The software you're using is using interpolation between samples when it's displaying the waveform, automatically.

Yes, but the original file is a perfect sine wave when played back, the one upsampled by Adobe Audition is too but the one upsampled linearly isn't. The program shows you what you actually hear. Linear connections between each sample don't represent the real world.

Quote:

Originally Posted by b0dhi /img/forum/go_quote.gif The spectral analysis seems accurate though. As expected, the linear upsampling mode measures poorly.

The artifacts you see in the spectral view are the linearly interpolated points. They add audible distortions. Why would you want to have these?

 

[b0dhi]

Quote:

Originally Posted by Hancoque /img/forum/go_quote.gif Yes, but the original file is a perfect sine wave when played back, the one upsampled by Adobe Audition is too but the one upsampled linearly isn't. The program shows you what you actually hear. Linear connections between each sample don't represent the real world.

No, the original is not a perfect sine wave. I just told you that the program is interpolating the waveform when it displays it. The dots are the sample points. The wavy lines in between, which make it look like a sine wave, are placed there by Audition and are not part of the original data.

Ofcourse, when the original file is played back it will look like a sine wave, because your DAC has a filter. The purpose of this thread is to avoid that sound.

Open the same file using Wavosaur (free), and take a look at what the original actually looks like.

 

[Hancoque]

Well, I said "the original file is a perfect sine wave when played back". But nonetheless you will get audible distortions using a linear interpolation. Why would you want these distortions, I ask again?

 

[b0dhi]

And I said "Ofcourse, when the original file is played back it will look like a sine wave, because your DAC has a filter. The purpose of this thread is to avoid that sound." The original file won't look anything like a sine wave when played back through a NOS DAC, for example.

The "distortions" you're referring to can only be called distortion if one assumes that the samples originally represented a sine wave.

 

[Hancoque]

Quote:

Originally Posted by b0dhi /img/forum/go_quote.gif The "distortions" you're referring to can only be called distortion if one assumes that the samples originally represented a sine wave.

And that's the case here.

There are two extreme cases:
1. You have pure square waves. Then linear interpolation is better because it preserves the sharpness of the edges.
2. You have pure sine waves. Then linear interpolation is worse because it doesn't properly reconstruct the sine wave.

In the first case the problem is that the 100% sharp edges are replaced by sine waves that only approximate the former shape.

In the second case the problem is that the inserted samples don't represent what would be there if the sample had been in the higher sample rate from the beginning.

Now the question is what is more audible. The first problem or the second problem. I would say the second problem is. You obviously would say the opposite.

Is that an accurate definition of what the whole discussion is about?

 

[Seidhepriest]

Anyway, please give a shout when the plugin's ready... It'd be rather fancy.

Just make sure it uses 32-bit internal transforms during upsampling and no sound gets hurt

By the way, a square-based instrument part was long a test track on a gear-testing CD. In some other places it might attract customers reacting to larger speakers shaking store.