lossyWAV 1.3.0 Delphi to C++ Translation Thread, Added noise linear PCM bitdepth reduction method. Options

[Nick.C]

Got sidetracked again - I'm going to add an option to convert any valid input audio (9-bit to 32-bit signed integer) to Float16, with or without adaptive / fixed noise-shaping. This will require the same messing about with the WAV handling code that I had to do for TransPCM. ETA is sometime early in the new year.

[Atak_Snajpera]

any progress on lossywav?

[Nick.C]

Still working on it, still distracted by the Elite: Dangerous KickStarter....

[FreaqyFrequency]

I've been experimenting with various compression levels in FLAC and -q X, and I've found that for certain tracks, -0 compresses much more efficiently than higher levels, and is practically equivalent with at least one of the tracks I tested at -8. Extraportable combined with -0 might very well make for the most overall efficient decode combination ever. Exciting stuff.

[Nick.C]

lossyWAV beta 1.3.0g attached to post #1 in this thread.

[FreaqyFrequency]

lossyWAV beta 1.3.0g attached to post #1 in this thread.

Neat! I'm getting an average of 2 to 3 kbps reduction at level 0 compression for the material I've tested so far. It also seems to average faster processing than 1.3.0a, which is what I was using previously.

[Nick.C]

I'm getting an average of 2 to 3 kbps reduction at level 0 compression for the material I've tested so far.

I've carried out a FLAC settings comparison test (based on my 55 problem sample test set) and will post the results this evening.

[Nick.C]

I did a bit of testing with my 55 problem sample set (12:43.933 of audio) processed using "-q X" and got the following results:

CODE

+------------------+---------------------+
| FLAC | Relative |
| Encoder +----------+----------+
| Setting | Size | Time |
+------------------+----------+----------+
| -8 -m -r 8 -p | 97.66% | 2187% |
| -8 -r 8 -p | 97.66% | 2089% |
| -8 -e -r 8 -p | 97.66% | 2094% |
| -8 -e -m -r 8 -p | 97.66% | 2105% |
| -8 -e -p | 97.66% | 1599% |
| -8 -p | 97.66% | 1615% |
| -8 -m -p | 97.66% | 1616% |
| -8 -e -m -p | 97.66% | 1634% |
| -6 -e -m -r 8 -p | 98.06% | 1689% |
| -7 -e -r 8 -p | 98.06% | 1694% |
| -6 -e -r 8 -p | 98.06% | 1695% |
| -7 -e -m -r 8 -p | 98.06% | 1697% |
| -4 -e -m -r 8 -p | 98.06% | 1701% |
| -7 -r 8 -p | 98.06% | 1705% |
| -7 -m -r 8 -p | 98.06% | 1710% |
| -5 -e -m -r 8 -p | 98.06% | 1723% |
| -5 -e -r 8 -p | 98.06% | 1799% |
| -7 -e -p | 98.06% | 1215% |
| -6 -e -p | 98.06% | 1220% |
| -6 -e -m -p | 98.06% | 1222% |
| -7 -m -p | 98.06% | 1222% |
| -7 -e -m -p | 98.06% | 1224% |
| -7 -p | 98.06% | 1225% |
| -5 -e -p | 98.10% | 983% |
| -5 -e -m -p | 98.10% | 1020% |
| -4 -e -m -p | 98.18% | 825% |
| -8 -r 8 | 98.36% | 368% |
| -8 -m -r 8 | 98.36% | 370% |
| -8 -e -m -r 8 | 98.36% | 374% |
| -8 -e -r 8 | 98.36% | 379% |
| -8 -e -m | 98.36% | 249% |
| -8 -m | 98.36% | 249% |
| -8 | 98.36% | 250% |
| -8 -e | 98.36% | 250% |
| -3 -e -m -r 8 -p | 98.69% | 1449% |
| -6 -e -m -r 8 | 98.80% | 294% |
| -7 -e -r 8 | 98.80% | 295% |
| -7 -r 8 | 98.80% | 295% |
| -5 -e -r 8 | 98.80% | 296% |
| -7 -e -m -r 8 | 98.80% | 296% |
| -6 -e -r 8 | 98.80% | 296% |
| -4 -e -m -r 8 | 98.80% | 298% |
| -7 -m -r 8 | 98.80% | 298% |
| -5 -e -m -r 8 | 98.80% | 318% |
| -6 -e -m | 98.81% | 174% |
| -7 -e -m | 98.81% | 175% |
| -6 -e | 98.81% | 176% |
| -7 -m | 98.81% | 177% |
| -7 -e | 98.81% | 183% |
| -7 | 98.81% | 189% |
| -3 -e -m -p | 98.82% | 605% |
| -5 -e | 98.86% | 127% |
| -5 -e -m | 98.86% | 138% |
| -4 -e -m | 98.95% | 105% |
| -3 -e -m -r 8 | 99.40% | 261% |
| -6 -m -r 8 -p | 99.42% | 201% |
| -6 -r 8 -p | 99.42% | 205% |
| -5 -r 8 -p | 99.42% | 206% |
| -5 -m -r 8 -p | 99.42% | 209% |
| -4 -m -r 8 -p | 99.42% | 215% |
| -6 -m -p | 99.42% | 137% |
| -6 -p | 99.42% | 138% |
| -5 -m -p | 99.45% | 106% |
| -5 -p | 99.45% | 108% |
| -4 -m -p | 99.51% | 87% |
| -3 -e -m | 99.55% | 71% |
| -3 -m -r 8 -p | 99.86% | 209% |
| -3 -m -p | 99.95% | 88% |
| -6 -m -r 8 | 99.97% | 28% |
| -6 -r 8 | 99.97% | 31% |
| -5 -m -r 8 | 99.97% | 32% |
| -4 -m -r 8 | 99.97% | 33% |
| -5 -r 8 | 99.97% | 33% |
| -6 -m | 99.97% | 6% |
| -6 | 99.97% | 7% |
| -5 -m | 100.00% | -1% |
+------------------+----------+----------+
| -5 | 100.00% | 0% |
+------------------+----------+----------+
| -4 -m | 100.06% | -5% |
| -3 -m -r 8 | 100.44% | 27% |
| -3 -m | 100.52% | -9% |
| -3 -e -r 8 -p | 100.54% | 678% |
| -3 -e -p | 100.69% | 279% |
| -3 -e -r 8 | 101.28% | 102% |
| -3 -e | 101.45% | 9% |
| -3 -r 8 -p | 101.76% | 78% |
| -3 -p | 101.87% | 20% |
| -3 -r 8 | 102.37% | -15% |
| -3 | 102.48% | -30% |
| -4 -e -r 8 -p | 102.59% | 846% |
| -4 -e -p | 102.74% | 403% |
| -4 -e -r 8 | 103.33% | 130% |
| -4 -e | 103.50% | 28% |
| -4 -r 8 -p | 104.08% | 85% |
| -4 -p | 104.19% | 22% |
| -4 -r 8 | 104.64% | -8% |
| -4 | 104.76% | -28% |
| -2 -e -m -r 8 -p | 108.34% | 113% |
| 0 -e -m -r 8 -p | 108.34% | 114% |
| 0 -e -m -r 8 | 108.34% | 114% |
| -2 -e -m -r 8 | 108.34% | 115% |
| -1 -e -m -r 8 | 108.34% | 115% |
| -1 -e -m -r 8 -p | 108.34% | 117% |
| -2 -e -r 8 -p | 108.34% | 118% |
| -2 -e -r 8 | 108.34% | 119% |
| -2 -r 8 | 108.50% | -4% |
| 0 -m -r 8 -p | 108.50% | -3% |
| 0 -m -r 8 | 108.50% | -3% |
| -1 -m -r 8 -p | 108.50% | -2% |
| -2 -r 8 -p | 108.50% | -1% |
| -2 -m -r 8 | 108.50% | -1% |
| -1 -m -r 8 | 108.50% | 0% |
| -2 -m -r 8 -p | 108.50% | 5% |
| -2 -e -m -p | 108.65% | -2% |
| -2 -e -m | 108.65% | -1% |
| -1 -e -m -p | 108.65% | -1% |
| 0 -e -m -p | 108.65% | -1% |
| 0 -e -m | 108.65% | -1% |
| -1 -e -m | 108.65% | 2% |
| -2 -e -p | 108.65% | 5% |
| -2 -e | 108.65% | 7% |
| 0 -m | 108.71% | -28% |
| 0 -m -p | 108.71% | -28% |
| -1 -m | 108.71% | -27% |
| -2 -m -p | 108.71% | -27% |
| -2 | 108.71% | -27% |
| -2 -m | 108.71% | -25% |
| -2 -p | 108.71% | -23% |
| -1 -m -p | 108.71% | -21% |
| 0 -e -r 8 -p | 110.94% | 30% |
| 0 -e -r 8 | 110.94% | 35% |
| 0 -r 8 -p | 111.12% | -31% |
| 0 -r 8 | 111.12% | -28% |
| 0 -e | 111.29% | -29% |
| 0 -e -p | 111.29% | -25% |
| 0 -p | 111.37% | -41% |
| 0 | 111.37% | -34% |
| -1 -e -r 8 -p | 113.83% | 34% |
| -1 -e -r 8 | 113.83% | 44% |
| -1 -r 8 -p | 114.06% | -27% |
| -1 -r 8 | 114.06% | -24% |
| -1 -e -p | 114.18% | -20% |
| -1 -e | 114.18% | -19% |
| -1 -p | 114.31% | -39% |
| -1 | 114.31% | -38% |
+------------------+----------+----------+

This post has been edited by Nick.C: Jan 23 2013, 22:20

[Nick.C]

lossyWAV beta 1.3.0h attached to post #1 in this thread.

[Nick.C]

lossyWAV beta 1.3.0i attached to post #1 in this thread.

[Nick.C]

lossyWAV beta 1.3.0j attached to post #1 in this thread.

[Cynic]

What is the penalty (if any) for using the -i switch? Slower encoding, higher risk of artifacts, or both? For someone not that concerned with filsesize, is the new noise shaping routine preferable to the default?

[Nick.C]

Hi Cynic,

The penalty is simply that the noise shaping is a bit more aggressive in trying to conform to the lossless audio spectrum. This saves about 1% of the resultant lossyFLAC size when compressed at --extraportable, less at higher quality levels.

I am currently working on a (very) simplistic psymodel based on masking using the Bark scale. This may be a wrong turn, but it is an option for both the classic method and the --interp-curve method of determining the basis shape of the noise spectrum.

Nick.

[Banned]

Out of curiousity, why does the distribution include a "gpl.txt" file? I thought it is for open source software.

[Nick.C]

When v1.40 is released, the source will be as well in a similar manner to the previous versions.

[Nick.C]

lossyWAV beta 1.3.0k attached to post #1 in this thread.

[edit] updated to beta 1.3.0k2 [/edit]
[edit2] updated to beta 1.3.0k3 [/edit2]

This post has been edited by Nick.C: Mar 8 2013, 22:26

[Nick.C]

lossyWAV beta 1.3.0m attached to post #1 in this thread.

[Nick.C]

lossyWAV beta 1.3.0n attached to post #1 in this thread.

[skamp]

Thanks for your work. I'm looking forward to 1.3.0 final.

[doccolinni]

I may be late for the party, but regarding the internal FFT routines...

You might consider using FFTW's genfft to generate efficient FFT routine source code. genfft is a set of routines written in OCaml which generate C code for efficient FFT routines. For example, this is a C code for 128-sample MDCT generated with genfft. These were used to generate FFTW's FFT routines, and you can use it to generate the code for lossyWAV's internal FFT routines. You can download it from here. Contact fftw@fftw.org and I'm sure they'll gladly assist you with using it to generate code for lossyWAV.

This post has been edited by doccolinni: Today, 03:11

[doccolinni]

By the way, you can read more about genfft and why FFTW is fast here.

The best solution, however, is to not have a specified internal FFT routine, because ultimately its performance will be hardware-specific (better/worse on some hardware). You are better off completely relying on FFTW since it adapts itself to whatever hardware it's used on.

This post has been edited by doccolinni: Today, 05:47