Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: lossyWAV Development (Read 573727 times) previous topic - next topic
0 Members and 3 Guests are viewing this topic.

lossyWAV Development

Reply #1100 –
Quote from: halb27 on
Quote from: Nick.C on
[Or, just allow the user to select a minimum-bits-to-keep between 0 and 8(?), defaulting to 3 for no user input?
That's ok for me, too.

Now that the encoder has changed a bit I'd like to do another listening test. Because listening tests aren't so much fun I'd like to do this at a time where the encoder is not expected to change again before the final release.
lossyWAV beta v0.9.6 attached to post #1 in this thread.
lossyWAV -q X -a 4 -s h -A --feedback 2 --limit 15848 --scale 0.5 | FLAC -5 -e -p -b 512 -P=4096 -S- (having set foobar to output 24-bit PCM; scaling by 0.5 gives the ANS headroom to work)

lossyWAV Development

Reply #1101 –
Guys, good work

I've been following this thread since its start, (tested it around 0.4 or so) and just thought to test it again.

I took a wav of a piece of song, encoded at q0, q5 and q10, and i really don't hear anything wrong at q0. (listening with headphones, volume near to top). Of course, this is not a direct ABX, but if i can't hear what to abx..

This song is noisy by design (reverb, distorted synths), so probably not the best one to hear for lossywav artifacts, but a proove of its usefulness.

The bottom side:

Flac -5 : 1017kbsp
lossywav -q 10 : 667kbps 561kbps
lossywav -q 5 : 499kbps 402kbps
lossywav -q 0 : 329kbps 289kbps

bottom line 2:
Just by curiosity, i encoded all them with lame 3.97 with -V 5 --vbr-new.
Original and -q 10 encode at the same bitrate, 144, while -q 0 encoded at 141.

[Edit: oops!!! I forgot the "-b 512" for flac.
 

lossyWAV Development

Reply #1102 –
Quote from: JAZ on
Guys, good work

I've been following this thread since its start, (tested it around 0.4 or so) and just thought to test it again.

I took a wav of a piece of song, encoded at q0, q5 and q10, and i really don't hear anything wrong at q0. (listening with headphones, volume near to top). Of course, this is not a direct ABX, but if i can't hear what to abx..

This song is noisy by design (reverb, distorted synths), so probably not the best one to hear for lossywav artifacts, but a proove of its usefulness.

The bottom side:

Flac -5 : 1017kbsp
lossywav -q 10 : 667kbps 561kbps
lossywav -q 5 : 499kbps 402kbps
lossywav -q 0 : 329kbps 289kbps

bottom line 2:
Just by curiosity, i encoded all them with lame 3.97 with -V 5 --vbr-new.
Original and -q 10 encode at the same bitrate, 144, while -q 0 encoded at 141.

[Edit: oops!!! I forgot the "-b 512" for flac.
Essentially, you're listening out for hiss as lossyWAV adds full spectrum noise when it removes bits from the samples.
lossyWAV -q X -a 4 -s h -A --feedback 2 --limit 15848 --scale 0.5 | FLAC -5 -e -p -b 512 -P=4096 -S- (having set foobar to output 24-bit PCM; scaling by 0.5 gives the ANS headroom to work)

lossyWAV Development

Reply #1103 –
Quote from: Nick.C on
Essentially, your listening out for hiss as lossyWAV adds full spectrum noise when it removes bits from the samples.

Nick, without slogging back through the previous 45 pages (I've read them all at one time or another, but not all tonight!) is there anything else specific we should be listening for at this point?

A little hiss isn't necessarily a bad thing. Analog tape is filled with it... and a vinyl groove can reproduce it nicely. If it was there in the beginning, and is too aggressively removed - as is all-too-often the case on modern reissues of classic material - the sound can be worse (which leads folks to spend all sorts of time tracking down earlier, un-remastered versions!).

    - M.

lossyWAV Development

Reply #1104 –
Apart from hiss there is a chance with the lower quality settings that the high frequency region sounds a tiny bit like having changed in pitch.
Last night I started my listening test, and with 00000_00595ms it's exactly like this when using -q 3.
(Not too much of a surprise though. IIRC AlexB provided this sample and found this very issue at a higher bitrate with a lossyWAV version several months ago).
The 'problem' at -q 3  is very subtle though as are the samples with added hiss.

In theory other problems may exist (any kind of distortion), especially with the very low quality settings, but we don't have any such experience so far.
lame3995o -Q1.7 --lowpass 17

lossyWAV Development

Reply #1105 –
I just finished my listening test with my usual problem samples Atemlied, badvilbel, bibilolo, 00000_00595ms, Blackbird/Yesterday, bruhns, dither_noise_test, eig, fiocco, furious, harp40_1, herding_calls, keys_1644ds, Livin_In_The_Future, S37_OTHERS_MartenotWaves_A, triangle-2_1644ds, trumpet, Under The Boardwalk.

I used -q 3 because this is a slightly lower quality setting than what was my transparency setting with the version I used with my last listening test.

My first 3 samples were 00000_00595ms, Atem-lied, and badvilbel. With 00000_00595ms I could hear the apparently changed pitch again and abxed it 7/10. With badvilbel I could here added hiss, arrived at 5/5 when abxing but missed afteerwards. Similar results for Atem-lied.
I do my listening test not only to find out about the goods and bads of lossyWAV in general but especially in order to find out which setting I should use with my real collection. With regard to this I am not content with -q 3 though I have to admit that my abx results aren't clear enough as a good basis for a decision. But I don't want to go so scientific: for my personal demands -q 3 isn't safe enough. Don't get me wrong: The deviations from the original are very subtle (to me, and my abx results show this).

So I tried -q 4, and this time I tried all my samples. Usually everything is fine, but I could abx the added hiss of badvilbel 8/10, and, as a surprise, triangle-2_1644ds 8/10. The problem with triangle is hard to describe: no hiss, no change in pitch, just some kind of very subtle distortion. I have the suspicion that S37_OTHERS_MartenotWaves_A (added hiss) and Under The Boardwalk (change in perceived pitch) aren't perfect either, but after a good start of 4/4 I missed badly.

I continued with -q 5 for these 4 samples. S37_OTHERS_MartenotWaves_A was okay now, but I abxed badvilbel and Under The Boardwalk 7/10. With triangle I arrived at 4/4 but missed later.

I am not as content with the current version as I was before.
While I think the quality is very acceptable at a quality level like -q 3 (only subtle issues) it's not like this for -q 5, at least not for me.

Do we have a regression? I'm afraid we have. I know listening tests in different situations aren't exactly comparable (at least my hearing abilities aren't always the same), and maybe the triangle problem existed before and I just didn't hear it.
But because I did several listening tests before which were more satisfying I'm afraid there is a regression.
The thing that changed recently as to my best knowledge was that the skewing was relaxed and the accuracy demands especially at the high frequency edge were strengthened (because of the general use of -1's spreading function). Maybe the high skewing was a good mechanism to take good care of the higher quality demands of problematic samples.
I'll try to investigate a bit in this direction.
lame3995o -Q1.7 --lowpass 17

lossyWAV Development

Reply #1106 –
Quote from: halb27 on
I am not as content with the current version as I was before.
While I think the quality is very acceptable at a quality level like -q 3 (only subtle issues) it's not like this for -q 5, at least not for me.

Strange. According to the helpfile q10 is highest quality and q0 is lowest bitrate, so most of the time chances are that q5 is better than q3 ?

Eleven steps in options are way too much for me at the moment. So when I aim for space saving I don't use parameters at all. With version 0.9.4 that equals to -0. A lossy image.flac resulted in 210 MB instead of 335 MB. Nice.
I noticed the progression to count up to 256 MB, then count down to 0, and then counting up again to the end which was 551 MB for that disc image. Savings 125 MB..

(I was trying to do so via mareo but somehow that failed. Will give it a try later. EAC > mareo > wav > lossywav > flac )

lossyWAV Development

Reply #1107 –
Quote from: collector on
Quote from: halb27 on

I am not as content with the current version as I was before.
While I think the quality is very acceptable at a quality level like -q 3 (only subtle issues) it's not like this for -q 5, at least not for me.

Strange. According to the helpfile q10 is highest quality and q0 is lowest bitrate, so most of the time chances are that q5 is better than q3 ? ...

Sure. What I tried to say is: with -q 3 (expected bitrate: ~335 kbps on average) my quality demands are such that I can accept the subtle deviations from the original. With -q 5 (expected bitrate: ~420 kbps on average) I personally don't though -q 5 quality sure is better than that of -q 3. With -q 5 I expect full transparency.
lame3995o -Q1.7 --lowpass 17

lossyWAV Development

Reply #1108 –
I tried again v0.8.8 which some time ago I found to be transparent at -4 with my samples.
Now I can hear the problems with badvilbel and Under the Boardwalk at -4. I didn't hear a problem with triangle but maybe I'm just less sensitive for this problem right now than I was this morning. After all it's a very subtle issue.
Going -2 I still could hear an increased hiss with badvilbel, but hear no problems with triangle and Under the Boardwalk.

So I think the main explanation is that right now I seem to be more sensitive towards the problems than I was before. It's not clear however whether apart from that there's a real quality advantage of v0.8.8 over v0.9.6.

More experience is very welcome.

P.S.: We shouldn't care too much about badvilbel. There's noise in the original, and a subtly added hiss onto it doesn't change a lot.
lame3995o -Q1.7 --lowpass 17

lossyWAV Development

Reply #1109 –
Quote from: halb27 on
.. with -q 3 (expected bitrate: ~335 kbps on average) my quality demands are such that I can accept the subtle deviations from the original. [..] With -q 5 I expect full transparency.

Thanks for your testing time and time again. From 0.9.6 we have a completely new quality scale, maybe your ideal setting has shifted to perhaps -q 5.6472  .
Is there a way to see what (internal) settings that are applied? nts, snr, bits_to_keep and such? To find a possible regression, the first thing would be to compare the parameter settings of the new version with those of a previous one.
In theory, there is no difference between theory and practice. In practice there is.

lossyWAV Development

Reply #1110 –
Because of my listening test results for triangle I also looked at it technically using -detail with v0.8.8 and v0.9.6.
Though the results aren't totally comparable it's hard to beleive that there should have been a regression with v0.9.6.
Guess differences heard are due to my different sensitivity this morning and this afternoon.
lame3995o -Q1.7 --lowpass 17

lossyWAV Development

Reply #1111 –
>> maybe your ideal setting has shifted to perhaps -q 5.6472 <<

Yes, obviously my ideal setting has changed, and for the biggest part I think it's due to an actual better hearing than the one I had when I did the listening tests before.

I just listened to the 4 critical samples using -q 6 and everything's fine.
So I will use -q 6 in the future - I don't have to care about a bitrate like 450 kbps.

But I suggest we change the default quality setting to -q 6 because we always wanted to have a transparent default setting.
lame3995o -Q1.7 --lowpass 17

lossyWAV Development

Reply #1112 –
Quote from: halb27 on
>> maybe your ideal setting has shifted to perhaps -q 5.6472 <<

Yes, obviously my ideal setting has changed, and for the biggest part I think it's due to an actual better hearing than the one I had when I did the listening tests before.

I just listened to the 4 critical samples using -q 6 and everything's fine.
So I will use -q 6 in the future - I don't have to care about a bitrate like 450 kbps.

But I suggest we change the default quality setting to -q 6 because we always wanted to have a transparent default setting.
Many thanks (yet again) for your efforts in listening to processed samples. Would it be better to:

a) Move current -q 6 to -q 5 stretching the higher presets and squeezing the lower presets;

or

b) Move all presets down one (adding a new -q 10 and -q 0 falls off the bottom).

Everything was going too smoothly....... .
lossyWAV -q X -a 4 -s h -A --feedback 2 --limit 15848 --scale 0.5 | FLAC -5 -e -p -b 512 -P=4096 -S- (having set foobar to output 24-bit PCM; scaling by 0.5 gives the ANS headroom to work)

lossyWAV Development

Reply #1113 –
I personally don't care much about it as long as the default is what is now -q 6.

Whether to drop current -q 0 or not depends on the usability of -q 0 with respect to what is to be expected by users of -q 0.
It would be kind if the one or other potential user of a low -q setting could share his opinion.
lame3995o -Q1.7 --lowpass 17

lossyWAV Development

Reply #1114 –
Quote from: Nick.C on
Many thanks (yet again) for your efforts in listening to processed samples. Would it be better to:

a) Move current -q 6 to -q 5 stretching the higher presets and squeezing the lower presets;

or

b) Move all presets down one (adding a new -q 10 and -q 0 falls off the bottom).

Everything was going too smoothly....... .


Oh, please, don't do that!  I do use -q 0, really! Please don't drop it!

But hey, you could do as vorbis does, adding something like -q -1
flac 1.2.1 -8 (archive) | aoTuVb5.7 -q 4 (pc, s1mp3)

lossyWAV Development

Reply #1115 –
Quote from: gasmann on
Quote from: Nick.C on
Many thanks (yet again) for your efforts in listening to processed samples. Would it be better to:

a) Move current -q 6 to -q 5 stretching the higher presets and squeezing the lower presets;

or

b) Move all presets down one (adding a new -q 10 and -q 0 falls off the bottom).

Everything was going too smoothly....... .
Oh, please, don't do that!  I do use -q 0, really! Please don't drop it!

But hey, you could do as vorbis does, adding something like -q -1
I'll squash and squeeze rather than remove the current -q 0.
lossyWAV -q X -a 4 -s h -A --feedback 2 --limit 15848 --scale 0.5 | FLAC -5 -e -p -b 512 -P=4096 -S- (having set foobar to output 24-bit PCM; scaling by 0.5 gives the ANS headroom to work)

lossyWAV Development

Reply #1116 –
ok, thank you! I would have been fine with q -1, too... As long as I can continue using it, it's alright

halb27 said users should share their opinion... well I regard myself a user  It's not transparent to me (I could just easily abx a song 7/7), but I like the fact that quality is much more stable than that of say mp3. I didn't find any serious problems on particular "problem samples". And this noise that is introduced is much less annoying than mp3 artifacts. Of course, at this bitrate mp3 generally does a better job, but I always have to fear there is a problem sample 

However, I don't use lossyWAV for archiving, that'll always have to be truely lossless, pardon. I use this low-bitrate flacs for listening only.
flac 1.2.1 -8 (archive) | aoTuVb5.7 -q 4 (pc, s1mp3)

lossyWAV Development

Reply #1117 –
I have two suggestions.

1.
Code: [Select]
-verbose
speaks for itself.

2.
To add to the lossyWAV metadata that gets saved in the wav files... the settings configuration string and version number of lossyWAV that were used to process the wav.

lossyWAV Development

Reply #1118 –
Quote from: gasmann on
... but I like the fact that quality is much more stable than that of say mp3. I didn't find any serious problems on particular "problem samples". And this noise that is introduced is much less annoying than mp3 artifacts. ...

Though I'm striving for transparency your description made me curious about the behavior of -q 0.
First I encoded my regular track set which I use for getting an idea of the average bitrate I have to expect when using a particular setting. The result was 263 kbps which is very low compared to what we had before with the lowest settings. The more was I surprised that I was pleased when listening to the encoded tracks. Quality is very good to me! This made me dare to use my problem samples with it. More surprise: abxing isn't very hard of course with most of the problems, but: with the exception of eig and furious the deviations from the original are not obvious at all and not at all annoying. Going -q 1 BTW (average bitrate: 281 kbps with my regular track set) made even furious not annoying to me and eig acceptable.

I didn't care much about the very low quality settings before, but, Nick, with your recent changes with the encoder I think you've succeeded in giving lossyWAV an extremely broad useful quality/bitrate range!
Thanks a lot.
lame3995o -Q1.7 --lowpass 17

lossyWAV Development

Reply #1119 –
Quote from: halb27 on
Quote from: gasmann on

... but I like the fact that quality is much more stable than that of say mp3. I didn't find any serious problems on particular "problem samples". And this noise that is introduced is much less annoying than mp3 artifacts. ...

Though I'm striving for transparency your description made me curious about the behavior of -q 0.
First I encoded my regular track set which I use for getting an idea of the average bitrate I have to expect when using a particular setting. The result was 263 kbps which is very low compared to what we had before with the lowest settings. The more was I surprised that I was pleased when listening to the encoded tracks. Quality is very good to me! This made me dare to use my problem samples with it. More surprise: abxing isn't very hard of course with most of the problems, but: with the exception of eig and furious the deviations from the original are not obvious at all and not at all annoying. Going -q 1 BTW (average bitrate: 281 kbps with my regular track set) made even furious not annoying to me and eig acceptable.

I didn't care much about the very low quality settings before, but, Nick, with your recent changes with the encoder I think you've succeeded in giving lossyWAV an extremely broad useful quality/bitrate range!
Thanks a lot.

I'll second that. I've been doing some testing with higher bit depths/sample rates (24/64, 24/88.2, 24/96). This was primarily to see if I could perceive any advantage to using them. As a starting point I decided to rip some tracks at 16/44.1 and encode tham at the lowest quality setting so that I could get a good idea of the type of degradation I was looking for. Very much to my surprise I could hardly hear a difference - just a very slight increase in "hiss" but at such a low level that I wouldn't have noticed it if I wasn't listening for it. I was expecting something like the hiss levels you used to get with cassette or a weak FM station.

On the "-q" settings. I don't have any particular axe to grind and don't want to muddy the waters but do we really need 10 quality levels? From lowest to highest we have a final bit rate of something like 250kbps to 550kbps so each change in -q setting gives only a very slight change in the result. Before LossyWAV came along I used Wavpack Lossy. I used the "-b" setting to set bits per sample rather than a specific bit rate. Using a BPS range of 3 to 6 gives pretty much the same final range in kbps as LossyWAV's 0 to 10 and I never found it inconvenient especialy since, like LossyWAV, it's possible to specify decimal number EG 4.6, 3.8 etc to get the result you want.

lossyWAV Development

Reply #1120 –
Quote from: Nick.C on
Quote from: halb27 on
I just listened to the 4 critical samples using -q 6 and everything's fine.
Would it be better to:

a) Move current -q 6 to -q 5 stretching the higher presets and squeezing the lower presets;

or

b) Move all presets down one (adding a new -q 10 and -q 0 falls off the bottom).

Or (too obvious?) just change the default to -q 6 for now?

I'd like to ask Halb27 if he's willing to do an ABX of (current) -q 5 vs.  -q 6 for those 4 problem samples. That would help rule out difference in hearing sensitivity.
In theory, there is no difference between theory and practice. In practice there is.

lossyWAV Development

Reply #1121 –
Quote from: botface on
On the "-q" settings. I don't have any particular axe to grind and don't want to muddy the waters but do we really need 10 quality levels? From lowest to highest we have a final bit rate of something like 250kbps to 550kbps so each change in -q setting gives only a very slight change in the result.

The more settings (11) the less steps I test. On my slow computer I skip the best and worst, so I test with 9,7,5 and 3 until I detect problems. Default is sufficient (whether 5 or 6).

lossyWAV Development

Reply #1122 –
Quote from: halb27 on
Quote from: gasmann on
... but I like the fact that quality is much more stable than that of say mp3. I didn't find any serious problems on particular "problem samples". And this noise that is introduced is much less annoying than mp3 artifacts. ...
Though I'm striving for transparency your description made me curious about the behavior of -q 0.
First I encoded my regular track set which I use for getting an idea of the average bitrate I have to expect when using a particular setting. The result was 263 kbps which is very low compared to what we had before with the lowest settings. The more was I surprised that I was pleased when listening to the encoded tracks. Quality is very good to me! This made me dare to use my problem samples with it. More surprise: abxing isn't very hard of course with most of the problems, but: with the exception of eig and furious the deviations from the original are not obvious at all and not at all annoying. Going -q 1 BTW (average bitrate: 281 kbps with my regular track set) made even furious not annoying to me and eig acceptable.

I didn't care much about the very low quality settings before, but, Nick, with your recent changes with the encoder I think you've succeeded in giving lossyWAV an extremely broad useful quality/bitrate range!
Thanks a lot.
I think that the -snr parameter has a lot to do with some of these problem samples.

I would propose something like

quality_signal_to_noise_ratios    : array[0..Quality_Presets] of Double  = (18,18.87,19.81,20.8,21.86,23,24.21,25.51,26.91,28.4,30);

instead of

quality_signal_to_noise_ratios    : array[0..Quality_Presets] of Double  = (16,17,18,19,20,21,22.8,24.6,26.4,28.2,30);
lossyWAV -q X -a 4 -s h -A --feedback 2 --limit 15848 --scale 0.5 | FLAC -5 -e -p -b 512 -P=4096 -S- (having set foobar to output 24-bit PCM; scaling by 0.5 gives the ANS headroom to work)

lossyWAV Development

Reply #1123 –
Quote from: Nick.C on
... quality_signal_to_noise_ratios    : array[0..Quality_Presets] of Double  = (18,18.87,19.81,20.8,21.86,23,24.21,25.51,26.91,28.4,30); ...

As this makes things more defensive: go ahead.
But why these strange steps like 18.87?
lame3995o -Q1.7 --lowpass 17

lossyWAV Development

Reply #1124 –
Quote from: halb27 on
Quote from: Nick.C on
... quality_signal_to_noise_ratios    : array[0..Quality_Presets] of Double  = (18,18.87,19.81,20.8,21.86,23,24.21,25.51,26.91,28.4,30); ...
As this makes things more defensive: go ahead.
But why these strange steps like 18.87?
I was looking for a smooth curve, so I worked out the power required to make 18 translate to 30 in 10 steps (i.e. snr:=power(snr[i-1],z)).
lossyWAV -q X -a 4 -s h -A --feedback 2 --limit 15848 --scale 0.5 | FLAC -5 -e -p -b 512 -P=4096 -S- (having set foobar to output 24-bit PCM; scaling by 0.5 gives the ANS headroom to work)