Best Free-software or no charge proprietary tool to downsample to 44.1 Options

[Morality124]

I have some audio files I've ripped that I want to down-sample to 44.1khz/16-bit format so they can burned to CDs. The audio files in questions are 24-bit (obv) and range from 96-192 kHz. I don't have any commercial proprietary audio software like Adobe Audition, but I do have the Free software programs Audacity and SoX at the moment, as well as EAC and ImgBurn. Would down-sampling using either of these programs be advisable, and if so, what is best settings and/or method(s) to do so? If not, what would be the recommended program and settings/methods? Thanks!

[skamp]

I use this:

CODE

sox -G "in.wav" -b 16 "out.wav" rate -v 44100 dither -a -s

See the manual for an explanation of the parameters and if you want to tweak them.

[Apesbrain]

If you have a number of files to convert you may find it easier to use foobar2000 with SoX resampler plugin.

[m45t3r]

The "Secret Rabbit Code" is a very good quality resampler and is completely Free Software (licensed as GPL). On the webpage you can found a plugin to mass convert files using foobar2000.

[Morality124]

Hmmm. I actually do have Foobar 2000 also, though I never thought of using it for this work (I also have the SoX plugin mentioned, but I was actually trying to use it for a different purpose). What exactly differs in quality between the methods mentioned?

[Morality124]

To clarify, what is the consequence of using a lower quality down-sampler? Abnormalities? A particular distortion?

This post has been edited by Morality124: Apr 29 2012, 07:53

[Juha]

If your files are type of WAV already then try Voxengo r8brain free.

Check some SRC Comparisons - http://src.infinitewave.ca/

Juha

[AndyH-ha]

There are some varying opinions on what is the "best" result for both resampling and for bit depth reduction. For bit depth this has mainly to do with the dither used. Unless something really gross is done, however, (and maybe even then) it is rather unlikely you could hear the difference.


For sample rate you can see some visual representations a
http://src.infinitewave.ca/
Look at the sweep tone conversion using Adobe Audition with filters and say Ableton Live 7. The Audition result is more ideal. There are other interesting displays. As with bit depth reduction, being able to hear a difference does not seem to be an established fact.

[stephan_g]

Hmmm. I actually do have Foobar 2000 also, though I never thought of using it for this work (I also have the SoX plugin mentioned, but I was actually trying to use it for a different purpose). What exactly differs in quality between the methods mentioned?

Anything of decent quality (I don't think people would bother with anything less here) should mainly differ in user experience. The only thing that may be potentially critical is clipping during downsampling - sometimes even hi-rez material is brickwalled and likely to require a level reduction.

In order to safeguard against this problem, I'd suggest the old SSRC "twopass" approach - downsample to 32-bit float first, determine peak levels, then dither to 16 bit.

Foobar should do all of that quite nicely with two converter runs and one Replaygain scan (multiple albums) in between, for just about as many files per pass as you like. I have verified that "32-bit" output does in fact produce float samples (v1.1.11). It's obviously helpful if Fb2k is set up to plainly show RG information in selection info, which makes it easy to determine total peak level.

If you find that downsampling produces levels in excess of 0.999969, the album(s) in question will require a small level reduction. That again can be taken care of with a converter run... enabling Replaygain in album mode with "prevent clipping" (only) should do the trick. Actually one could integrate that with converter run #2 (dithering to 16 bit), now that I think about it - it would only affect material that would clip otherwise, and no manual inspection of peak levels would be required.

[Arnold B. Kruege...]

There are some varying opinions on what is the "best" result for both resampling and for bit depth reduction. For bit depth this has mainly to do with the dither used. Unless something really gross is done, however, (and maybe even then) it is rather unlikely you could hear the difference.


For sample rate you can see some visual representations a
http://src.infinitewave.ca/
Look at the sweep tone conversion using Adobe Audition with filters and say Ableton Live 7. The Audition result is more ideal. There are other interesting displays. As with bit depth reduction, being able to hear a difference does not seem to be an established fact.

It's interesting to see how CEP/Audition which used to get no respect among pros and semi-pros always did a lot of things right, even when more "pro", prestigious, and expensive software like some old releases of Pro Tools didn't even do the basics right.

This post has been edited by Arnold B. Krueger: Apr 30 2012, 14:40

[stephan_g]

It's interesting to see how CEP/Audition which used to get no respect among pros and semi-pros always did a lot of things right, even when more "pro", prestigious, and expensive software like some old releases of Pro Tools didn't even do the basics right.

Some of those look quite scary indeed. In some cases you have to wonder how much of an understanding of SRC the programmers had.

I suppose it's a case of "black box syndrome". Musicians are used to tools that do something which has to be determined empirically. Verification of function against a spec is not usually an issue. Now obviously, an SRC's function is pretty well-defined and verifiable, as with a speaker amplifier. I guess that has yet to sink in. Hence, people are basically "flying blind" and questioning of performance (negative feedback) rarely takes place.

This post has been edited by stephan_g: May 2 2012, 09:59

[phofman]

SoX results on http://src.infinitewave.ca are very decent, considering the price and availability :-)

[rednyrg721]

Audacity 2.0 results are much worse than Audacity 1.39 results for some reason.

[stephan_g]

That one puzzled me, too. I think I'll ask the devs...

EDIT: Ah, so it seems the 1.3.9 results were obtained using libsamplerate, while 2.0 apparently uses libresample (whose "high quality" setting isn't a match for libsamplerate's "best" by a long shot). It looks like there were some issues with libsamplerate, like this one and this. Seems like a Linux thing... I'd have to check how the Windows build is configured. ... The same, it seems. Hmm. I don't think I like that.

EDIT^2: Ah, I think that's the problem:
Resample.h

CODE

    libsamplerate, written by Erik de Castro Lopo.  GPL.  The author
    of libsamplerate requests that you not distribute a binary version
    of Audacity that links to libsamplerate and also has plug-in support.

And as we all know, Nyquist and VST plugins are supported by default (VST was added after 1.3.9 then I guess, can't remember). So yet another "private build only" feature (next to ASIO with its proprietary license)... *sigh*

Guess I'll reconfigure that and update my build.

This post has been edited by stephan_g: May 2 2012, 17:40

[john33]

If you're Windows based and like libsamplerate, you could try srcdrop on the 'Others' page at Rarewares.

[lvqcl]

EDIT^2: Ah, I think that's the problem:

This issue is also mentioned at http://wiki.audacityteam.org/wiki/Libresample

[stephan_g]

If you're Windows based and like libsamplerate, you could try srcdrop on the 'Others' page at Rarewares.

'fraid I'm thoroughly spoiled by the Foobar2000 / SoX resampler combo (best quality AND ease of use)...

(Before that, I'd typically use SSRC. The ringing around filter cutoff makes it a less-than-ideal choice if sample rates significantly below 44.1 are involved, but other than that it still performs well.)

This issue is also mentioned at http://wiki.audacityteam.org/wiki/Libresample

I think I even read that one back when I first compiled Audacity, but - not being aware of the quality implications at the time - had all but forgotten about it.

[Morality124]

If you're Windows based and like libsamplerate, you could try srcdrop on the 'Others' page at Rarewares.

'fraid I'm thoroughly spoiled by the Foobar2000 / SoX resampler combo (best quality AND ease of use)...

If I use SoX from the command line (which I have for de-emphasis, stereo channel flipping, and inverted absolute phase), will this produce the same results as the method you are using?

[bandpass]

They're essentially the same. Architectural differences between the standalone version and the Foobar port might result in some variation in the least-significant bit of the converted audio, but such differences are negligible and can and do occur for other reasons too.

Skamp's sox command line suggestion above should produce good results. (Though you could probably save a little CPU time by dropping the -v option—it's really intended for cases when both input and output audio is at least 24-bit.)

[phofman]

If I use SoX from the command line (which I have for de-emphasis, stereo channel flipping, and inverted absolute phase), will this produce the same results as the method you are using?

The foobar sox plugin is code taken from the sox rate effect and modified to fit foobar plugin API. I think I saw in the foobar plugin source code there was support for modern CPU directives added too (SSE, etc.) I do not think I saw that in the original sox sourcecode (it is kept portable to other CPU architectures). I did not check if the results are identical.

If sox rate.c code gets modified in the future (for whatever reason), the foobar plugin functionality can start deviating. That does not mean any problem at all, just an information.

This post has been edited by phofman: May 3 2012, 12:48