I have a 24 bit, 96 khz wav source.
I was thinking of putting it on a dvd; to do this I would have to reduce to 24 bit/48 khz
I was going to make the dvd using DVD-Lab pro, where wav files are played during menus;
and each track has a menu that goes to the next track when it ends.
Is anyone familiar with using dvd-lab pro menu/wav?
Will the dvd player process 24-bit 48 khz audio and output through rca cables?

Last thing; The source is 5 channel flac; one channel is LFE.
I'm going to convert to stereo using foobar. This should work?

This is my first post on the forum after reading lots of threads

What is the source material? If you're going to reduce it to stereo anyhow, why not consider simply transferring to CD? The difference should be inaudible except in certain crazy edge cases that don't represent normal listening.

The source is Pink Floyd DSOTM original master which was remastered to 4.1.
I was doing some listening tests with 24bit/44khz and 16bit/44khz wav samples from a 24bit/96khz source and I could tell which one was 24 bit.

CD is only 16-bit. I might make a 24 and a 16 bit just for convenience.

May be it is better to use sox for conversions?

Thanks ZinCh.
something like this with sox to convert to cd specs wav?

sox input.wav -b 16 output.wav rate -h 44100 dither

Out of curiosity, how did you prepare said samples? And how were the listening tests performed? The difference should not be audible when done properly.

Continuing on from Canar's post...

Can you provide an ABX report and samples to see if there are others who can distinguish a difference as well?

sox input.wav -b 16 outfile.wav rate 44100 dither -s

you can also use it with multichannel flac input, something like:

sox -c 6 input.flac -b 16 -c 2 outfile.wav rate 44100 dither -s

(-c number of channels for input/output)

The tests were not good tests, as I didn't use dithering on the conversion to 16-bit.






I was using Mangione05.wv from the other thread. I converted that in foobar (again, no dithering) Really my tests only confirmed that I can tell the difference between 24-bit and 16-bit without dither.

New test with 2 tracks.
1st track: 24/96
2nd track: 16/44

track 2 was converted from track 1 using sox with dither

I copied them many times randomly and mixed them around- then dragged them in a bunch to foobar with no signs (bitrate, etc) visible. I listened to a total of 37 tracks and marked down + or - for 24 and 16 respectively
I identified 22 out of 37 correctly. That's 59.46% success rate.

(trk#) -- (1st listen) / (2nd listen)---------------(actual track properties)

1--XX/16...............16
2--24/24................24
3--16/16................16
4--24/16................24
5--16/24...........,....16
6--24/16................16
7--24/16................16
8--16/24................24
9--24/16...........,,...24
10--16/24.........,....16
11--24/24..............24
12--24/16........,.....16
13--16/24........,.,...24
14--24/24.........,....24
15--16/16.........,....16
16--24/16.........,....24
17--16/24..............16
18--16/16..........,...24
19--24/24..............16

This result is guessing. Use the ABX tool in Foobar and try again.

Not so good. If you were flipping coins, you'd do only 9.46% worse.

To put this into perspective you need to compare your results to a like number of coin flips.

Here's a tool for doing that:

http://stattrek.com/Tables/Binomial.aspx

This page also explains many of the reasons why.

I plugged your results in there and got a "cumulative probability" of about 0.90

Here's their definition of cumulative probability:

"A cumulative binomial probability refers to the probability of getting AT MOST a specific number of successes in a specific number of trials. For instance, we might ask: What is the probability of getting AT MOST 2 Heads in 3 coin tosses. That probability (0.875) would be an example of a cumulative binomial probability"

IOW the probability of getting at most 22 right out of 37 tries is about 0.90. There's about 1 chance in 10 of doing what you did by dumb, pure luck. The usual guideline for testing is that you want your cumulative probability at the 0.95 or 0.99 level. You are way short of that.

I've done a lot of testing and it is not that unusual to get scores in the 0.90 range.

If you think there is a chance of better results, run a test with a ton more trials and see what happens. The more trials you run, the lower percentage of right answers will be required for acceptable results.

Usually what happens is that after running more trials your results converge asymptotically to 50% right, and you have compelling evidence that you were guessing all along.

OTOH you may be onto something. Don't fret, don't fume, either buy into the data you have obtained so far, or make more! ;-)

I would also like to hear about your test enviroment - audio interface, transducers, etc.