Hi.

Welcome to HA. I sincerely hope your experience here is as enlightening and meaningful as mine has been.

Regards,
Madrigal

I fail to understand the b value of 128 on -V 0. Is there a simple explanation for why the bitrate should not vary in the full 32-320 kbps range? Also, can -b 32 override -V 0?
For example would running:
-V 0 -b 32 -B 320
make any difference at all, or would the preset-style V switch override the b and B?

I believe the -b 32 will override the default (128 in this case).

/Add: I think the simple explaination why 32-112 range should not be used at all is that more or less any samples encoded with these bitrates will not be transparent (except digital silence - which is btw encoded as 32kbps anyway).

On a resembling matter, I've been using Audiograbber (ugh!) with the lame_enc.dll (3.96.1), with -V 2 equvalent, and it appears that it uses -b 96 as default? Should be 128. It could be that Audiograbber sets -b 96 internally, but it seems unlikely. I ask because I wonder if the lame dll has the same mappings as the exe.

you shouldn't mess with the --presets at all. So just use -V0 and that's it. No -b or -B unless you like to waste space or worse sound quality
Anyway, LAME 3.96.1 uses -b 128 as default, however the newest LAME 3.97 alpha builts use -b 32. That's why it is the best to just stick with the presets. Note: LAME Alpha and Beta release shouldn't be used for anything else than testing!

The way I understand it is this: lame has different variable bitrate modes. you have ABR and a true VBR mode. (the ABR is what was used with the old r3mix.net settings)
However, apparantly the VBR mode is somewhat unreliable below 112 Kbits/sec. It will mostly produce good encodes but may output some low-quality parts. Therefore, lame in VBR mode (which is what the presets use) has a bitrate floor of 112 Kbits/sec.

Though I may be wrong

No, --r3mix was VBR, not ABR.

.
Not anymore. At least --vbr-new mode in the current 3.97 alpha.

Hi!

I'm new to this forum. I registered because I needed some encoding information for my new-born Creative MuVo TX SE. Since it's only 256MB, I'd like to fill it up with proper encodings.

Sorry if this is the wrong section or something. I tried searching for topics about mp3 player encodings but failed.

So. I use dBPowerAmp which has command line access to Lame, and Lame is the latest 3.97b I believe. The one recommended here anyway.

Since I'm probably going to invest to Sennheiser PX-100 or something similar, I figured -V 6 might be appropriate.

-V6 --vbr-new was in mind actually. Now I have read about this -Y parameter, which cuts everything above 16kHz. I tried it, but it just produced the same results as in normal -V6. Also, I checked the full command line reference, and it was not there. Is this somehow obsolete and/or replaced by some other parameter?

I put my money on --highpass 16 , but I dont think it did anything. Probably not the switch I'm searching for or completely wrong anyway.

I'm not an Audiophile. I was considering -V8 but thought that new headphones might require the -V6 so I was sticking with that. I guess its pretty good.

I can now fit 3 -V2 albums on this player. More is required and that quality is an overkill for me anyway (although I do encoded all my CD's with that).

Thank you for your time!

And the final question is this: The whole thing with --alt preset being ABR, was this true in some versions? I always wanted the recommended --alt preset standard but ended up with bitrates never topping 224kbps. With the latest Lame though, no problems, just curious about why did I had this experience in the past.

A quote from http://lame.sourceforge.net/doc/html/switchs.html:
"--highpass highpass filtering frequency in kHz
Set an highpass filtering frequency. Frequencies below the specified one will be cutoff."

-V6 --vbr-new uses a lowpass of 15600 Hz. You should not add any other switches. In my opinion -V6 --vbr-new is a good and balanced option to use with portables. I have found that it produces bitrates from about 100 to 150 kbps with various music genres (http://www.hydrogenaudio.org/forums/index....ndpost&p=335491).



In my experience -V8 is inferior. It resamples to 32 kHz and uses a lowpass of 12500 Hz, but that doesn't even solve the quality problems -V8 has below 12500 Hz. However, just the lowpass setting clearly changes the perceived quality with many music types if the listener can hear frequencies over 12500 Hz. On the other hand, hearing a lowpass of 15.600 Hz is much harder. You can try that for example by encoding a test sample with -b 320 and -b 320 --lowpass 15.6. (I have done personal ABX tests and I suppose other listening tests that can confirm this have been published at HA.)


Edit: There seems to be a small typo in the LAME documentation: "Set an highpass filtering frequency."

I might be a bit thick, but whats the point using --lowpass ?

I tried -b 320 --lowpass 12.5 just for kicks, and the filesize was the same, however the sound was clearly inferior.

I need to mentiont that I never understood these kHz terms. Sometimes they represent how high a sound is and sometimes... not? Tried also --lowpass 15.6 and it sure was transparent to the original.

That's a CBR encoding, obviously lowpassing has no effect on the filesize...

Cutting some more or less inaudible high frequencies off can make the overall audible quality better. The space is always limited with MP3 files. LAME adjusts this automatically so you don't have to worry about manual tweaking.

I mentioned -b 320 just because it is the highest quality setting LAME 3.97b1 has. It is always constant 320 kbps so the resulting file size is also constant like Gambit said. Using it for testing the audibility of a lowpass setting would minimize the possible effect of other factors.

So I shall leave that setting alone for portable use. -V6 then is the way to go

Is there an updated copy of the chart in the first post that would coincide with LAME 3.97b? For example, in my own observations I've found that -V3 now uses a lowpass around 16000.

Please let me know if and where I can find a current reference. Thanks!

I have posted an updated table here: http://www.hydrogenaudio.org/forums/index....ndpost&p=335491. The table was for b1, but I think b2 uses the same lowpass values.

Thanks. I appreciate the update.

That's because -V 3 and greater use -Y switch.

the minimum bitrate should be removed from the chart since it is no longer being used

Sorry for possible offtopic. I just found this topic as closest to my problem, but it can equally well be a new topic.

So, what's the reason for me to be confused. When I've changed my SB Live! 5.1 to Audigy 2 ZS, I just started to hear some higher frequencies of sound (including my huge collection of self-grabbed mp3s). And I used (and use LAME to encode them.

I started with YAMP encoder which was using LAME 3.80, then I moved to EAC & LAME 3.92 (newest release for that time). I've been hearing and analyzing the sound spectre of output mp3s, and arrived at a conclusion that the best (in size/quality) variant for me is:


Then there was 3.93 and 3.93.1. I've been grabbing and compressing and expanding my mp3 collection. Grabbing, and compressing, and extending...

Then here came the 3.95.
Right away I noticed that mp3s I'm getting with it (using same parameters) are oftenly slightly smaller. That was the reason to REGRAB and RECOMPRESS all my collection (about 40 or 50 GB for that time).

Then there were 3.95.1, 3.96 and 3.96.1... Regrabbing, recompressing, extending collection...

And then "here came a new challenger" (© Toshiden Battle Arena . I've bought Audigy 2 ZS.

First time there was some kind of euphoria - I started to hear HIGHER sounds on my computer! All that hi-hats' and cymbals' high harmonies, and so on...

Then it turned to some kind of depression. I started to hear mp3-specific distortions in MY MP3s! Which were made with my LAME parameters (I also recommended using them to all my friends... and so on). It was a real shock for me, because I used to consider my mp3s ideal.

So, I started to re-examine various mp3 codecs, incl. FhG, GoGo, SCMPX, Blade, Xing and recently appeared LAME 3.97.

The results I got for LAME shocked my once more. Awful, just AWFUL distortions all above 16-17 KHz. Even 320 at Stereo (not Joint Stereo) mode, and even with filter disabling (-k). I tried lot of different options and their combinations, all of them resulted much worse quality then trivial 3.93.1 at 320, Stereo mode (with -k), or at VBR2 Stereo (with -k also), and even worse then LAME 3.93.1 of GoGo gives me at 224 kbps or even 192 kbps (no matter Stereo or Joint Stereo mode)!

I tried using q0, and V0, and so on - it didn't improve anything enough.

I also tried various mp3 decoders: Winamp 5.1, Adobe Audition (in which I analyze sound frequency spectre) and LAME by itself. In Audiotion results were even more awful, then in Winamp and LAME, but in them both spectre was still too bad at look (at freqs above), and I heared that all in Winamp.

Even more - those distortions (looking like some kind of collapse of frequency spectre graph) grew more and more for 3.95.1 -> 3.96.1 -> 3.97. I just did't beleive my eyes that LAME produced that collapse.

So I temporary turned back to 3.93.1, which produces an ideal graph allover the sound spectre (even on freqs above 20 KHz!) when using the following parameters:

The resulting mp3 freq graph doesn't drift in any significant way from the from original WAV graph. 320 kbps' graph (@3.93.1) also doesn't differ from the original in any way, but average filesize increases significantly.

Could your please explain, what happened with LAME starting from 3.95? (or 3.94, which I didn't test)
Could it be some problem of my decoder (even despite I used LAME to decode mp3s)? Or what?

Thank your in advance.

P.S. I can post, or send, or attach here (if it's allowed), or put at your disposal in any other way the graphical results of spectral analysis I carried out.

Spertrograms are no judge of sound quality. Use an ABX comparsion tool to make sure you hear differencies between the original file and the compressed mp3.

(edit- spelling)

-k will absolutely destroy the quality of the files. Also, M/S stereo will make things sound WORSE. Leave it as joint stereo (it is NOT lossy in LAME). -h does nothing for VBR, -q0 does nothing for VBR. Setting min/max bitrate is not needed in 3.97 when using VBR.
Spectral analysis is virtually useless for measuring sound quality.


Stop screwing with the parameters and just use -V2 --vbr-new
Nothing more, nothing less. Try LAME 3.97 or 3.98a7 if you prefer going a little on the wild side. Then ABX it if you still think you hear "distortions".

-m s isn't such a bright idea either.

Never mind, you caught it in your edit.

Thanx for your support, people

But, as far as I understood, ABX test is based on "blind" listening to audio, but my acoustic system (and perhaps my ears ))) is not ideal, so in future (when I purchase a better one) I could regret my choise made now. Are there any less subjective, MORE objective test? Which can reveal the real (no only heared) differences between two soundtracks?

My father offered my to "subtract" original signal from the encoded->decoded one, and then to estimate the "difference signal" remained. Is this method good, how do you think? Are there any software to do that? AFAIK, Sound Forge and CoolEdit can't...

Or may be some graphic tests?

And one more question. You say spectral analysis couldn't be objective in this way. But if it shows some evident distortions, collapses etc - could they NOT be real distortions of sound and its quality? Especially if they aren't above 20 KHz, but within 16-20 KHz interval?

(Of course, I agree that even identical graphs do not obligatory mean identical sound, or sound of identical quality.)

P.S. By the way, I did't understand from the FAQ, what does -vbr-new key?

P.P.S. Joint Stereo "pushes together" channels, when they spead stronger than in "regular song"... Am I wrong? Does Stereo mode produce better quality? Does it increase filesize considerably?

To me the main problem is that you seem to think it's absolutely necessary to get everything until 20 kHz.
However our ears are less and less insensitive the more you go beyond 16 kHz. This is true even for young people though they are able to hear stuff beyond 16 kHz.

If you care so much about that you should absolutely do an abx test. You don't need good equipment to find out whether or not you need HF beyond say 18 kHz.
Try Lame -V3 which has an 18 kHz lowpass and report us about whether or not you're really missing HF or find things distorted (which of course is possible only if you can abx something).

You should be a lot more afraid of -k which might give you more HF but as well a much higher chance that you get audible distortions.

Unlike your ears, your eyes aren't meant for hearing.

No.

Acutally CoolEdit can, but it tells you absolutely nothing about how your brain interprets what your ears are hearing.

NO! Listening tests are all that matter.

Of course they represent real "distortions", but you need to ask yourself if you can hear these distortions in a blind listening test.

Read this.

No, it usually produces worse quality at any given bitrate.

For the same quality? Yes!

CoolEdit can do that easily, but...

Our ear is extremely complex. "Visual" differences, even "extremely evident" visual differences... may not be audible. Why? Simply because our hearing, whcih comprises the ears, the inner ear system, the auditory nerves, and the brain, will perform some extremely complex post-processing and filtering.

Case in point: Try encoding with the highest quality of LAME ( -V 0), or Vorbis (-q 10). Decode. Subtract from the original wave. You will see "extremely evident" visual differences. But even if you pump the decoded wave (i.e. not original) through a 1-million-dollar universal-audiophile-grade equipment... the chances that you will hear any difference is less than 1 ppm (part per million).

Like I said above, what's visible to your eye may never, ever be audible to your ear.

Spectral analysis is not un-objective, it is simply useless. "Visible" distortion, even "extremely evident" visible distortion, like I wrote above, may not be audible to your ear even the slightest.

You are partially wrong.

In low-bitrate, Joint Stereo is really Intensity Stereo. *That* will collapse stereo separation.

On higher bitrates, Joint Stereo is actually Mid-Side (MS) Coding. Instead of coding L & R channels separately, it encodes M=(L+R)/2 and S=(L-R)/2. The L & R channels can be recovered very easily: L=M+S and R=M-S. The reason for using MS is that since the S channel usually requires far less bitdepth, encoding using MS will be smaller than encoding using LR, provided that the difference between L & R are not too big. LAME is capable of determining when to use MS and when to use LR.

See this page: http://harmsy.freeuk.com/mostync/ (also contained in Greynol's link to "Joint Stereo" on HA wiki).

Edit: Rats. Greynol is faster

"(The mere fact that Joint Stereo is used in lossless compression ought to be enough to destroy - in one stroke - the myth that JS "destroys stereo separation")" (cite from http://harmsy.freeuk.com/mostync/) - I like this argument! It's quite cogent But then I don't understand for what reason most CBR mp3s are made using S mode (even on LAME). And for what reason all other Stereo modes are supported by LAME?


CoolEdit can subtact noise, but it's not the same. It uses the noise sample not "as is", but as a kind of "spectre", am I wrong? Or did I miss some function of CoolEdit, or Adobe Audition (for now)?


But LAME 3.93 and GoGo didn't produce them... Is this "feature" (new to 3.95 and the following versions) just an improvement - i.e. to give more bits to the "more useful" freqs?

...So, as I understood, the only way for me to get a better quality is to get a better soundsystem and to carry out the ABX test.

By the way, can you suggest how to choose most appropiate sample for tests? I like metal music, and oftenly there are lot of live cymbals & hihats, which sound good, and have much high harmonies...
...Or it just may be any sample where I can hear distortions (or maybe just invent them ?

Because of the fallacy that S mode (or shall we say, LR encoding) is better than JS mode (or shall we say, MS encoding).

Because in cases where L & R channels are extremely different, MS encoding is not better than LR encoding. But because the S channel is usually allocated less bitdepth, it may cause degradation.

Here's how (IIRC - I don't have Audition on my office computer):
1. Open first wave.
2. Invert.
3. Select all.
4. Copy.
5. Open second wave.
6. Paste mix.

You simply use the Mix Paste function (Ctrl+Shift+V), but again, it tells you nothing about what a lossy file sounds like. Pepoluan already mentioned this, but you also need to make sure your files are synchronized since lossy files often introduce delay.

I'm not sure what you're seeing or what you're not seeing but it is utterly pointless to judge how something sounds based on what it looks like.

The more bits used trying to reproduce high frequencies which you will not be able to hear mean fewer bits that are available to reproduce what it is that you can hear. Encode something with Blade @ 192 and the same thing with Lame @ 192, I don't care what versions you use; look at them and then listen to them.

The ABX test in this case is supposed to help you determine whether you can actually tell the difference between the original source and a lossy version of it. I think it's commonly accepted that a high-end sound system is not necessary to accomplish this.

Just use the music that you ordinarily listen to. Samples with lots of cymbals and hi-hats are a good way to reveal some of the more easily identifiable artifacts.

Do you have foobar2000 installed on your system with the ABX component?

No, I downloaded foobar2000, but didn't find any ABX component for it (using search on its site). Is abchr I downloaded a little earlier worse for some reason?

By the way, using abchr, I couldn't hear those artifacts I see... It was only one test as now, but maybe the sample I used wasn't suitable. I'll try another samples and/or another sound system (and/or someone else's ears ).

I also tried to mix paste (thanx Pepoluan and Greynol for that idea ) samples using Adobe Audition, but it was too difficult to guess how great is the "lossy offset", mentioned by Greynol. Simple comparison of files' lenght did not help me. Is there any bright idea how to find out that lossy offset?

As I see, EAC can determine the offset generated by specific codec, but this operation returns an error for some reason (I tried to determine offset for GoGo and two different LAMEs). What can be a reason? I didn't find anything about it at EAC website's FAQ.

And some more questions about notorious spectral analysis. )

1) for two same-sounding (for me ) samples, is the one with "nicer" spectre (i.e. most bordering upon original) a better one (concerning quality)? Or it is no judge of quality even for same-sounding samples?

2) For two samples encoded using the same original, do same-looking graphs signify the same quality? (of course, accurate within my eyes )

And more questions... (It would be great grateful (edited ) if the LAME developer will answer )
1) Why qval=3 is now default (at least, when using -V2 --vbr-new)? As I remember it was qval=2 in older LAME versions...
2) What is --vbr-new?
3) Are there any differences between two spellings: -V 2 (readme file provided with LAME 3.97) and -V2 (found here, on hydrogenaudio.org)?
4) Please tell me, does -k really "destroy the quality of files" (like Firon said)? Almost any LAMEs mp3 graph remains same-looking when using this key, and the only difference is lowpass-filter-collapse dissapperance. Or that means nothing again?..


But isn't JS (when using at LAME, on high bitrates) mostly M/S stereo? Or your mean forced M/S, without using L/R Stereo when it's needed?

It is included as part of the installation. I don't remember if it gets selected by default.

From what I understand, this test is designed so that people could rank different files rather than determine whether they could tell the difference between two files.

It sounds like abchr was able to accomplish the same thing as abx from the point of view that you realized that you couldn't tell the difference.

You need to switch to Multitrack View and load the two files one under the other, zoom in to a portion of the audio that you can easily distinguish. Drag one of the files so they line up to the precise sample and then lock them. Right click on one file and choose edit. Then select the entire track. Switch back to multitrack view again. Right click on the other file and choose edit. Copy what is currently selected. Switch back to multitrack view. Right click on the first file again. Use the mix-paste function and check that you want to invert both channels.

It's a pretty outdated and IMO useless feature. Anyway, I wouldn't bother with this at the moment.

What you hear is all that matters. It is nice to see what lossy codecs throw out but please give up on this method of determining quality.

Please give up on this method of determing quality.

Thanx a lot, you all just broke down my hasty and wrong conclusions.

But I still would be gratefull if someone answers my remaining questions.

2) The new VBR method, faster and reputed several times to be of equal or better quality than the old.
3) No.
4) Well, you're choosing to keep additional frequencies which you probably cannot hear and which the MP3 format has some difficulties encoding. And yes, sight graphs mean very little in the context of audio!

1 - because it makes no difference for VBR NEW, note: LAME 3.98 qval will default to 0, but it still makes no difference!
2 - it's a different approach for VBR encoding, we started development some years ago, it will be the default VBR mode for LAME 3.98
3 - no, with single letter options taking parameters you can leave the blank out
4 - -k is a shortcut for setting high-/low-pass filters to minimum/maximum. LAME 3.98 will not have this switch anymore, but you can still change lowpass as before.

Wow, LAME 3.98 should be interesting! Thanks for the information, robert!

Thanks Dv1989 and thanks Robert (an answer from LAMEs developer is a nice thing)!


But why an old algorythm is default for now? Is the new one still a "flaw design"?


What's about single letter? -V 2 and -V2 is what I'm interested about. Dv1989 said that's all the same...

Before 3.97 development started --vbr-new wasn't widely used though it was available before (called 'fast' mode then).
With 3.97 (where I think has been work done with --vbr-new) it was found --vbr-new is of comparable quality to --vbr-old. Sometimes it's better, sometimes it's worse, and in the overall view there seems to be a tendency that is't slightly better. AFAIK we don't have a very solid basis to prefer one over the other qualitywise, but as --vbr-new is faster it is preferable to use --vbr-new.

Within the same version I think it's wise not to change the defaults, and you can use whatever you like by own settings.

3.98 will be a major step ahead, vbr behavior is significantly improving, and --vbr-new is default with 3.98.

The developers were obviously not confident enough, or did feel that it made sense, to have LAME default to the new mode. As we have just discovered, this has changed by now! Ooh, it's exciting stuff.


Robert meant the same as I: you can leave out the space between the switch (in this case -V) and the parameter (i.e. 2) - and the end result will be the same.

Thanx for all! I'll try using -V2 --vbr-new, and I think it will be good for me.

Hmm... But when not setting minimum frame bitrate, LAME uses bitrate lower than 128 (on some songs - oftenly)... is it good?

Between the reservoir and complexity of the frames to be encoded, less than 128 kbits can still provide transparency.

Not in LAME though. LAME doesn't have Intensity Stereo (unless it got it recently...), but FhG and probably others do.



I'm sorry. I meant forcing L/R stereo (-m s) for all frames.


Yes, because it determined that it didn't need 128 for that particular frame to reach the quality level you specified.

Thanks again Now I see, LAME is even more clever than I thought a few days ago.
I like it, and I like this forum and its members who helped me so much.

I'm trying to use EAC with LAME and have tried to use what seems to be a common and correct command line parameter - but I'm receiving the following error mesage:


The external compressor returned an error!

Options: -V 0 --vbr-new

I've tried this with multiple CDs and receive the same error.

The LAME files are in the EAC folder and that location is specified in EAC setup...

Try removing the space from between the "V" and the "0".

You may also want to try "-V1 --vbr-new" and "-V2 --vbr-new". I very much doubt if you'll hear any difference but the files will be much smaller. "-V3 --vbr-new" suits me just fine, but everyone's equipment and ears are different.

Cheers, Slipstreem.

The space between the V and the 0 is perfectly fine. I'm thinking %s and %d have been left out.

Please read the wiki instead of resurrecting a dead topic.

Sorry - you were right. Actually I did read the Wiki - probably 30 times along with many other web sites that explained how to configure. I did NOT realize tha the %s and %d were necessary - actually couldn't find what those tags did at all - and they weren't listed as part of the instructions from other setup instructions.

Clearly, however, they must be important because when I added them in, everything seemed to start working fine...

So sorry for not understaning the instructions better - but thanks for pointing me in the right direction.

Thanks!

Wow, this is getting really off-topic, but here...
http://www.exactaudiocopy.de/en/index.php/...sion-questions/