Difference between revisions of "LossyWAV"

From Hydrogenaudio Knowledgebase
Jump to: navigation, search
m
Line 4: Line 4:
 
| caption =  
 
| caption =  
 
| maintainer = [http://www.hydrogenaudio.org/forums/index.php?showuser=42400 Nick.C]
 
| maintainer = [http://www.hydrogenaudio.org/forums/index.php?showuser=42400 Nick.C]
| stable_release = v0.6.7 RC2
+
| stable_release = v0.9.2 RC3
| preview_release = beta v0.9.1
+
| preview_release =  
 
| operating_system = [[Wikipedia:Microsoft Windows|Windows]]
 
| operating_system = [[Wikipedia:Microsoft Windows|Windows]]
 
| use = [[Wikipedia:Digital signal processing|Digital signal processing]]
 
| use = [[Wikipedia:Digital signal processing|Digital signal processing]]
Line 63: Line 63:
 
!10 Album Test Average
 
!10 Album Test Average
 
| 850kbps
 
| 850kbps
| beta v0.7.9
+
| v0.9.2 RC3
| 496kbps
+
| 479kbps
| 453kbps
+
| 425kbps
| 403kbps
+
 
| 375kbps
 
| 375kbps
 
| 352kbps
 
| 352kbps
| 333kbps
+
| 334kbps
 
| 318kbps
 
| 318kbps
 +
| 306kbps
 
|-
 
|-
 
!53 sample "problem" set                 
 
!53 sample "problem" set                 
 
| 784kbps  
 
| 784kbps  
| beta v0.7.9
+
| v0.9.2 RC3
| 561kbps
+
| 543kbps
| 518kbps
+
| 494kbps
| 462kbps
+
| 433kbps
| 434kbps
+
 
| 408kbps
 
| 408kbps
 
| 385kbps
 
| 385kbps
 
| 365kbps
 
| 365kbps
 +
| 348kbps
 
|}
 
|}
  
Line 185: Line 185:
 
===Application settings===
 
===Application settings===
 
<pre>
 
<pre>
lossyWAV beta v0.9.1, Copyright (C) 2007,2008 Nick Currie.
+
lossyWAV v0.9.2 RC3, Copyright (C) 2007,2008 Nick Currie.
 
lossyWAV is issued with NO WARRANTY WHATSOEVER and is free software.
 
lossyWAV is issued with NO WARRANTY WHATSOEVER and is free software.
  
Line 194: Line 194:
 
Quality Options:
 
Quality Options:
  
-1/1a/1b      quality preset 1, circa 4??kbps for 44.1khz, 2ch;
+
-1/1a/1b      quality preset 1, circa 480kbps for 44.1khz, 2ch;
-2/2a/2b/2c    quality preset 2, circa 4??kbps for 44.1khz, 2ch [default];
+
-2/2a/2b/2c    quality preset 2, circa 425kbps for 44.1khz, 2ch [default];
-3/3a/3b/3c    quality preset 3, circa 3??kbps for 44.1khz, 2ch;
+
-3/3a/3b/3c    quality preset 3, circa 375kbps for 44.1khz, 2ch;
-4/4a/4b/4c    quality preset 4, circa 3??kbps for 44.1khz, 2ch;
+
-4/4a/4b/4c    quality preset 4, circa 350kbps for 44.1khz, 2ch;
-5/5a/5b/5c    quality preset 5, circa 3??kbps for 44.1khz, 2ch;
+
-5/5a/5b/5c    quality preset 5, circa 335kbps for 44.1khz, 2ch;
-6/6a/6b/6c    quality preset 6, circa 3??kbps for 44.1khz, 2ch;
+
-6/6a/6b/6c    quality preset 6, circa 320kbps for 44.1khz, 2ch;
-7/7a/7b/7c    quality preset 7, circa 3??kbps for 44.1khz, 2ch;
+
-7/7a/7b/7c    quality preset 7, circa 305kbps for 44.1khz, 2ch;
 
               [a/b/c = 1/2/3 extra FFT analyses; default=3,2,2,2,2,2,2(max=5)]
 
               [a/b/c = 1/2/3 extra FFT analyses; default=3,2,2,2,2,2,2(max=5)]
  
Line 217: Line 217:
 
-correction  write correction file while processing WAV file; default=off.
 
-correction  write correction file while processing WAV file; default=off.
 
-merge        merge existing lossy.wav and lwcdf.wav files.
 
-merge        merge existing lossy.wav and lwcdf.wav files.
-shaping <n>  enable fixed noise shaping using n as a proportion of full noise
+
-scale <n>    scaling factor from WaveGain / etc; default = 1.000000.
              shaping, i.e. 0=off, 1=fully on, 0.5=half on, etc.; default=0.00 .
+
              (incompatible with -autoshape)
+
-autoshape    enable variable noise shaping (depends on bits-to-remove);
+
              (incompatible with -shaping)
+
-scale <n>    scaling factor from WaveGain / etc.; default = 1.000000.
+
  
 
System Options:
 
System Options:
Line 237: Line 232:
 
David Robinson for the method itself and motivation to implement it.
 
David Robinson for the method itself and motivation to implement it.
 
Don Cross for the original Pascal source for the FFT algorithm used.
 
Don Cross for the original Pascal source for the FFT algorithm used.
Horst Albrecht for valuable tuning input and feedback.
+
Horst Albrecht for valuable tuning input and feedback.</pre>
Sebastian Gesemann for helping me to implement noise shaping properly.
+
</pre>
+
  
 
===Example Foobar2000 converter settings===
 
===Example Foobar2000 converter settings===

Revision as of 20:40, 13 April 2008

lossyWAV
{{{logo}}}

Developer(s) Nick.C
Release information
Initial release {{{released}}}
Stable release v0.9.2 RC3
Preview release
Compatibility
Operating system Windows
Additional information
Use Digital signal processing
License GNU LGPL
Website Hydrogenaudio

lossyWAV is a new free lossy pre-processor for PCM audio contained in the WAV file format. It reduces bit depth of the input signal, which, when used in conjunction with certain lossless codecs, reduces the bitrate of the encoded file significantly compared to unpreprocessed compression. lossyWAV's primary goal is to maintain transparency with a high degree of confidence when processing any audio data.

History

lossyFLAC is an idea started by 2Bdecided at Hydrogenaudio, utilising the wasted bits feature of the FLAC lossless codec with the aim of transparently reducing audio bit depth (making some lower significant bits (LSB's) zero), consequently taking advantage of FLAC's detection of consistently-zeroed lower significant bits within each single frame and significantly increasing coding efficiency.[1] In this way the user can enjoy audio encoded using the same codec (which may be all important from a hardware compatibility perspective) at a reduced bitrate compared to the lossless version.

Nick.C ported the original MATLAB implementation to Delphi (Many thanks CodeGear for Turbo Explorer!!) with a liberal sprinkling of IA-32 and x87 Assembly Language for speed.

Subsequently, lossyFLAC proved itself to work with other lossless codecs, so the application name was changed to lossyWAV.

Since then, Nick.C has heavily developed and built upon lossyWAV, with valuable tuning performed by halb27 at Hydrogenaudio.

Indicative bitrate reduction

It must be stressed that lossyWAV is a pure variable bit rate pre-processor. Bits-to-remove from the audio data are calculated on a block-by-block basis (codec-block length = 512 samples, 11.6msec @ 44.1kHz) using overlapping fast Fourier transform (FFT) analyses of at least two lengths (default quality preset (-2) = 64, 256 & 1024 samples). After some manipulation, the results of each FFT analysis for a specific codec-block are then grouped and the minimum value used to determine bits-to-remove for the whole codec-block. Bit removal adds noise to the output, however the added noise has been pre-calculated and its level will be at or below the noise floor of the codec_block in question. Each sample in the codec-block is then rounded such that the first <bits-to-remove> lsb's are zero. In this way the wasted bits feature of FLAC et al is exploited.

lossyWAV Test Set FLAC -8 Version lossyWAV -1 lossyWAV -2 lossyWAV -3 lossyWAV -4 lossyWAV -5 lossyWAV -6 lossyWAV -7
10 Album Test Average 850kbps v0.6.7 RC2 494kbps 450kbps 402kbps
53 sample "problem" set 784kbps v0.6.7 RC2 558kbps 515kbps 462kbps
10 Album Test Average 850kbps v0.9.2 RC3 479kbps 425kbps 375kbps 352kbps 334kbps 318kbps 306kbps
53 sample "problem" set 784kbps v0.9.2 RC3 543kbps 494kbps 433kbps 408kbps 385kbps 365kbps 348kbps

File identification

lossyWAV-processed WAV files are named with a double filename extension, .lossy.wav, to make them instantly identifiable. e.g. ".lossy.flac" would indicate an audio file which was processed using lossyWAV, and subsequently encoded using FLAC.[2]

From beta v0.6.1, the -correction parameter is used when processing to create a correction file which is named with the .lwcdf.wav double filename extension. When "added" to the corresponding .lossy.wav, using the -merge parameter, the original file will be reconstituted.

Combinations of lossyWAV with each specific encoder are referred to as lossyX, where X is an abbreviation of the lossless codec name. Combination names are listed in the "known supported codecs" section below.

From beta v0.5.9, lossyWAV inserts a variable-length FACT chunk into the WAV file immediately after the FMT chunk. This takes the form:
fact/<size>/lossyWAV beta vx.y.z : dd/mm/yyyy hh:mm:ss
-2 -cbs 512 -nts 0.00 -snr 21.00 -skew 36.00
-spf 22224-22235-22336-12347-12358 -fft 10101
Where the version, date & time and user settings are copied. Additionally, if a lossyWAV FACT chunk is found in a file, the processing will be halted (exit code = 16) to prevent re-processing of an already processed file.

The -check parameter can be used to determine whether a file has previously been processed without trying to process it, exit code = 16 if already processed; exit code = 0 if not.

Quality presets

  • -1: Highest quality preset, disc space-saving alternative to lossless archiving for large audio collections.
  • -2: Default preset; a compromise between -1 and -3.
  • -3: High quality preset for usage on a compatible DAP, approx. 400kbps for "normal" music. [3]

All tuning has been performed on quality preset -3 with -2 and -1 being more conservative. Quality preset -3 is generally accepted to be (and from testing so far is) transparent. If you find a track which -3 fails to achieve transparency after processing, please post a sample (no more than 30 seconds) in the development thread.

Apart from the quality presets the -nts (noise threshold shift) parameter is the most important parameter to control quality. Without noise threshold shifting (-nts 0) the number of bits to be removed is computed in a theoretically optimal way. -nts 0 is defaulted when using quality preset -3.

In order to be defensive a negative -nts value can be chosen up to -48. -nts -2 is defaulted when using quality preset -2, and -1 defaults to -nts -4. For archiving purposes and/or very cautious users even more conservative values may be of interest.

Because of internal precautions in addition to 2Bdecided's principles experience so far tells us that a small positive -nts value keeps the encoding transparent or brings up only subtle differences. This way file size can be decreased. A -nts value of more than 10 however is not recommended.

Supported input formats

  • WAV: 4-bit to 32-bit integer; sample rate ≥ 32kHz PCM. Very high sample rates (>48kHz) have not been extensively tested. Tunings have been focussed on 16-bit, 44.1kHz samples (i.e. CD PCM).

Codec compatibility

Codec Supported Encoder parameters Combination name
Apple Lossless No
FLAC Yes -5 -b 512 --keep-foreign-metadata lossyFLAC
LA No
LPAC Yes -b512 lossyLPAC
Monkey's Audio No
MPEG-4 ALS Yes -l -n512 lossyALS
OptimFROG No
TAK Yes -fsl512 lossyTAK
TTA No
WavPack Yes --blocksize=512 lossyWV
WMA Lossless Yes lossyWMALSL

There is also evidence — so-called "Bit Shifting" — to suggest that lossyWAV may work with MLP, but this remains untested due to prohibitive prices of encoders.

A comparison of portable media players is here, which shows FLAC and WMA Lossless compatibility among listed players. Any player supported by Rockbox can use FLAC or WavPack files after installing Rockbox.

Using lossyWAV

Application settings

lossyWAV v0.9.2 RC3, Copyright (C) 2007,2008 Nick Currie.
lossyWAV is issued with NO WARRANTY WHATSOEVER and is free software.

Usage   : lossyWAV <input wav file> <options>

Example : lossyWAV musicfile.wav

Quality Options:

-1/1a/1b       quality preset 1, circa 480kbps for 44.1khz, 2ch;
-2/2a/2b/2c    quality preset 2, circa 425kbps for 44.1khz, 2ch [default];
-3/3a/3b/3c    quality preset 3, circa 375kbps for 44.1khz, 2ch;
-4/4a/4b/4c    quality preset 4, circa 350kbps for 44.1khz, 2ch;
-5/5a/5b/5c    quality preset 5, circa 335kbps for 44.1khz, 2ch;
-6/6a/6b/6c    quality preset 6, circa 320kbps for 44.1khz, 2ch;
-7/7a/7b/7c    quality preset 7, circa 305kbps for 44.1khz, 2ch;
               [a/b/c = 1/2/3 extra FFT analyses; default=3,2,2,2,2,2,2(max=5)]

Standard Options:

-o <folder>   destination folder for the output file(s).
-nts <n>      set noise_threshold_shift to n dB (-48.0dB<=n<=+36.0dB);
              (-ve values reduce bits to remove, +ve values increase).
-snr <n>      set minimum average signal to added noise ratio to n dB;
              (0.0dB<=n<=48.0dB) Increasing value reduces bits to remove.
-force        forcibly over-write output file if it exists; default=off.
-check        check if WAV file has already been processed; default=off.
              errorlevel=16 if already processed, 0 if not.
-noclips      set allowable number of clips per channel per codec block to 0;
              default= -1=0,-2=1,-3=2,-4=3,-5=4,-6=5,-7=6.
-correction   write correction file while processing WAV file; default=off.
-merge        merge existing lossy.wav and lwcdf.wav files.
-scale <n>    scaling factor from WaveGain / etc; default = 1.000000.

System Options:

-quiet        significantly reduce screen output.
-nowarn       suppress lossyWAV warnings.
-detail       enable detailed output mode

-below        set process priority to below normal.
-low          set process priority to low.

Special thanks:

David Robinson for the method itself and motivation to implement it.
Don Cross for the original Pascal source for the FFT algorithm used.
Horst Albrecht for valuable tuning input and feedback.

Example Foobar2000 converter settings

Foobar2000 Converter Settings.PNG

Example flossy3.bat file called from Foobar2000

@echo off
z:\bin\lossyWAV %1 %3 %4 %5 %6 %7 %8 %9 -below -nowarn -quiet
z:\bin\flac.exe -5 -f -b 512 "%~N1.lossy.wav" -o"%~N2.flac" 
del "%~N1.lossy.wav"

Frequently asked questions

  • Question: Is it VBR?
  • Short answer: Yes.
  • Question: Is it transparent?
  • Short answer: Almost certainly.
  • Question: Is it lossless?
  • Short answer: No.
  • Question: Will it ever have a CBR mode?
  • Short answer: No.
  • Question: Why should I use this?
  • Answer:
  • high quality
  • extremely low chance of audible artefacts
  • reasonable bitrates
  • usable with unmodified, established lossless formats.

External links