Here's my finding after playing with the LAME (3.92) encoder:
Fixed bit rate: 128, 160, 192, 320
Presets:
--alt-preset standard.mp3
--alt-preset extreme.mp3
--alt-preset insane.mp3
Note: I didn't play around with noise shaping etc, I could drive myself nuts with all the combinations.
This is better! I like this a lot better than the Fraunhofer encoder built into iTunes. At all bit rates, the Lame decoder sounds like at least one bit rate higher in the Fraunhofer decoder. But it has a better handle on transients like cymbal splash and piano notes at all rates.
However, you don't get something for nothing, with all that compression there are still artifacts. What interesting to me is that the variable bit rate artifacts are different for LAME than VBR Fraunhofer where the fixed bit rate artifacts in both methods are more or less the same.
Piano notes and their decay were noticeably effected even at the highest bit rates. As were the transients from plucked acoustic bass notes. As bit rate increases, they become less and less noticeable.
With LAME's ability to handle transients better, I'd have a hard time at the higher bit rates being able to pick mp3 vs. original in a blind test. I think I can still pick them pretty well at 128 kbps, at 192 kbps I'd guess, I'd be missing a few. At 320 kbps I'd be better than random but close.
This is with a very revealing system which I'm familiar with, and music which I know well. I can't imagine doing a bind test on this stuff with a marginal unfamiliar system and unfamiliar music. Those famous MP3 tests are crap for that reason.
Sheldon
Fixed bit rate: 128, 160, 192, 320
Presets:
--alt-preset standard.mp3
--alt-preset extreme.mp3
--alt-preset insane.mp3
Note: I didn't play around with noise shaping etc, I could drive myself nuts with all the combinations.
This is better! I like this a lot better than the Fraunhofer encoder built into iTunes. At all bit rates, the Lame decoder sounds like at least one bit rate higher in the Fraunhofer decoder. But it has a better handle on transients like cymbal splash and piano notes at all rates.
However, you don't get something for nothing, with all that compression there are still artifacts. What interesting to me is that the variable bit rate artifacts are different for LAME than VBR Fraunhofer where the fixed bit rate artifacts in both methods are more or less the same.
Piano notes and their decay were noticeably effected even at the highest bit rates. As were the transients from plucked acoustic bass notes. As bit rate increases, they become less and less noticeable.
With LAME's ability to handle transients better, I'd have a hard time at the higher bit rates being able to pick mp3 vs. original in a blind test. I think I can still pick them pretty well at 128 kbps, at 192 kbps I'd guess, I'd be missing a few. At 320 kbps I'd be better than random but close.
This is with a very revealing system which I'm familiar with, and music which I know well. I can't imagine doing a bind test on this stuff with a marginal unfamiliar system and unfamiliar music. Those famous MP3 tests are crap for that reason.
Sheldon
Nice you took the time and invented into playing around with
lame. Looks like you are real a pro at it now!
May it be possible to offer or send me a short test clip,
lossless compressed (flac for example) with the part of the sample
showing the most obvious problem for --alt-preset standard?
I spent much time with testing this preset and welcome every
sample that really messes it up.
regards
lame. Looks like you are real a pro at it now!
May it be possible to offer or send me a short test clip,
lossless compressed (flac for example) with the part of the sample
showing the most obvious problem for --alt-preset standard?
I spent much time with testing this preset and welcome every
sample that really messes it up.
regards
Not really. What like to do with LAME (built into CDex) is set the VBR from 32 to 320 kb/s and to New and the quality to 0 (best). This way, the MP3 doesn't waste space when high bitrate is not required, like in passages without highs or lows, and when they are present, it uses up only as much as needed. I did a blind test on myself between a .wav and an .mp3 of the same song encoded in this way. I could VERY barely tell the difference, but the filesize was about hald as large as with a 320kb/s CBR.
the first calculation of the bit rate of CDs was correct @ 705.6 kbps, you just forgot about the fact that there are 2 tracks on a stereo recording. So if you multiply 705.6 x 2 you get 1411.2kbps the total bit rate of a CD.
Dont have time atm to read all the posts in this thread but I dont understand where all the MP3 bashing comes from really.
I have a good system so if MP3's are really to sound rubbish then id hear it. I use a computer as a transport with uncompressed Wavs from CD's. The comps better at this then my TEAC T1 transport so dont tell me the comps not got a good enough digiout.
To say you cant hear a difference between MP3 and wavs is daft, a wav always sounds better then the MP3, I did a few comparisons. But to say the MP3 is not good is daft too. I get lots of enjoyment out of my MP3's most are 128 or above and I dont pine for better quality when I do. They dont sound thin or lacking in anything and the trebble quality is fine too.
I have a good system so if MP3's are really to sound rubbish then id hear it. I use a computer as a transport with uncompressed Wavs from CD's. The comps better at this then my TEAC T1 transport so dont tell me the comps not got a good enough digiout.
To say you cant hear a difference between MP3 and wavs is daft, a wav always sounds better then the MP3, I did a few comparisons. But to say the MP3 is not good is daft too. I get lots of enjoyment out of my MP3's most are 128 or above and I dont pine for better quality when I do. They dont sound thin or lacking in anything and the trebble quality is fine too.
Referring to the original layer 3 trials, the average score was in the range of 4.5, with 4 being excellent, and 5 being indistinguishable from the original signal.
Now, for my two cents:
Listening tests I have performed on the Stax Signature system, with a decent-but-noisy bitstream dac, indicates that (with the music I sampled) an MP3 of 256kbps or higher is equivalent to the original signal, if you don't explicitly try to discover errors, and are not used to pinpointing MP3 distortion. Lower bitrates were clearly distinguishable. At all bitrates I tried (up to 320kbps), listening fatigue set in at some point, which it did not do with the original signal.
Using the MusePack (MPC) encoder, which (amongst other things) does not set its lowpass at 16kHz, the quality 8 setting (averages 220kbps) is usually indistinguishable from the original, unless the material plays tricks with the phase, such as the dog on Three Wishes (which btw sounds marvellous on the Stax Signatures, due to the excellent low frequency reproduction). I have not yet experienced listening fatigue from MusePack.
I've not yet pinpointed the patterns to look for in MusePack, nor do I intend to try, since I am actually happy listening to it.
However, in MP3, the patterns will be rather clear if you listen to a certain piece (preferrably a clear one, with few instruments and very high detail) at low bitrates, and then higher bitrates, until you hit 256kbps. Do that a couple of times, and you will be able to identify the patterns in 320kbps materials. I've not tried >320kbps, but expect the same will be the case there.
MP3 is less offensive in-room than in-head. YMMV; if you're not a earspeaker person, you may already have trouble with the soundscape-in-your-head effect that might interfere with the listening itself.
One thing I would suggest that someone try, if we've got any MP3 software coders here, is to introduce a delay to signals whose phase is negative. Our ears are based on absolute-phase sensitive resonators, that do not get excited by the first negative halfcycle of a signal, or at least not until at least 5ms later on. If you introduced this filter into the encoder, you would probably greatly improve the perceived lack of ambience.
Loss patterns are interesting. If you listen to a minidisc, on certain pieces of music, it will seem as though (visualized) the sound is built up of 2D layers at different depths, and you can suddenly get an angled view, revealing the flaws in a most annoying manner. This description is probably rather vague, though. 😉
Anyway..
Just so noone gets the wrong idea, I am certainly in favour of trying to find more compact ways of transporting sound to the consumer. I just don't think we've gotten there yet.
There is also the argument that simplifying the sound, by removing components that are inaudible (just make sure they are inaudible to everyone, on all equipment, not just on poor equipment, or on unaware subjects), should improve reproduction.
Certainly, having a simpler waveform to reproduce will allow amplifiers and speakers to perform better. This may be the source of some of the characterizations as "better sounding". IMD distortion, for one, should be reduced significantly.
And, to clarify again, MP3 at high bitrates does sound good. Just not as good as I'd like. MPC is almost there.
Now, for my two cents:
Listening tests I have performed on the Stax Signature system, with a decent-but-noisy bitstream dac, indicates that (with the music I sampled) an MP3 of 256kbps or higher is equivalent to the original signal, if you don't explicitly try to discover errors, and are not used to pinpointing MP3 distortion. Lower bitrates were clearly distinguishable. At all bitrates I tried (up to 320kbps), listening fatigue set in at some point, which it did not do with the original signal.
Using the MusePack (MPC) encoder, which (amongst other things) does not set its lowpass at 16kHz, the quality 8 setting (averages 220kbps) is usually indistinguishable from the original, unless the material plays tricks with the phase, such as the dog on Three Wishes (which btw sounds marvellous on the Stax Signatures, due to the excellent low frequency reproduction). I have not yet experienced listening fatigue from MusePack.
I've not yet pinpointed the patterns to look for in MusePack, nor do I intend to try, since I am actually happy listening to it.
However, in MP3, the patterns will be rather clear if you listen to a certain piece (preferrably a clear one, with few instruments and very high detail) at low bitrates, and then higher bitrates, until you hit 256kbps. Do that a couple of times, and you will be able to identify the patterns in 320kbps materials. I've not tried >320kbps, but expect the same will be the case there.
MP3 is less offensive in-room than in-head. YMMV; if you're not a earspeaker person, you may already have trouble with the soundscape-in-your-head effect that might interfere with the listening itself.
One thing I would suggest that someone try, if we've got any MP3 software coders here, is to introduce a delay to signals whose phase is negative. Our ears are based on absolute-phase sensitive resonators, that do not get excited by the first negative halfcycle of a signal, or at least not until at least 5ms later on. If you introduced this filter into the encoder, you would probably greatly improve the perceived lack of ambience.
Loss patterns are interesting. If you listen to a minidisc, on certain pieces of music, it will seem as though (visualized) the sound is built up of 2D layers at different depths, and you can suddenly get an angled view, revealing the flaws in a most annoying manner. This description is probably rather vague, though. 😉
Anyway..
Just so noone gets the wrong idea, I am certainly in favour of trying to find more compact ways of transporting sound to the consumer. I just don't think we've gotten there yet.
There is also the argument that simplifying the sound, by removing components that are inaudible (just make sure they are inaudible to everyone, on all equipment, not just on poor equipment, or on unaware subjects), should improve reproduction.
Certainly, having a simpler waveform to reproduce will allow amplifiers and speakers to perform better. This may be the source of some of the characterizations as "better sounding". IMD distortion, for one, should be reduced significantly.
And, to clarify again, MP3 at high bitrates does sound good. Just not as good as I'd like. MPC is almost there.
MP3 quality is acoustically indistinguishable from the CD if done properly. One of the key mistakes people make is to use CBR or even ABR, and also "stereo" thinking it will give better channel separation.
Listen to an MP3 ripped using EAC then encoded joint-stereo VBR1 from 32-320 using LAME. Amazing quality.
If you TRULY believe you can tell the difference: Take a single song that you know well from a new cd, rip it using CBR from 128-320 then do a VBR recording as set up above. Put them all in winamp, no shuffle, no repeat. Choose Misc-> shuffle (but DO NOT look at the song order). Play the songs through as many times as you'd like... write down what you think each song is... go back and check.
I will BET you'd be surprised.
Listen to an MP3 ripped using EAC then encoded joint-stereo VBR1 from 32-320 using LAME. Amazing quality.
If you TRULY believe you can tell the difference: Take a single song that you know well from a new cd, rip it using CBR from 128-320 then do a VBR recording as set up above. Put them all in winamp, no shuffle, no repeat. Choose Misc-> shuffle (but DO NOT look at the song order). Play the songs through as many times as you'd like... write down what you think each song is... go back and check.
I will BET you'd be surprised.
Why is the MP3 format still being discussed?
There are much better lossy formats out there now like OGG Vorbis and MPC.
There are much better lossy formats out there now like OGG Vorbis and MPC.
mp3 is the most widely supportet. There are so many media out there that can handle mp3.
mpc has near to nothing to use it outside your PC.
vorbis doesn´t sound better at high bitrates
mpc has near to nothing to use it outside your PC.
vorbis doesn´t sound better at high bitrates
Well, I'm not going to touch the "mp3's sound perfect vs mp3's sound awful" debate, as nothing useful will come out of arguing over individual preferences.
I would, however, like to tell everyone about a very interesting alternative codec. It's called FLAC, for Fully Lossless Audio Codec. And it is exactly that - completely lossless, meaning that there is NO lost information - nothing whatever is thrown away from the original .wav file, but through the magic of compression algorithms, compression is still possible. This is similar to the way you can zip a text file, compressing its size, but without losing a single character contained in the file when you finally unzip it.
FLAC does not produce the huge compression ratios that mp3 does - in my experience, the output .flac file is about one-third the size of the input .wav file. That means you can compress the average CD to about 200 MB instead of the original 600 MB. That still means
you can put about 5 CD's of material on one gig of hard drive space, or about one thousand CD's on a single 200 Gigabyte drive.
With the rate at which hard-drive and flash-memory capacity is increasing, in a very few years many of us will be deciding that it wasn't worth throwing away some audible quality for the sake of compressing our CD-quality audio files into mp3's. When that happens, most people will have to re-rip their entire collection. If you use FLAC, there won't be a need to do that: you can convert FLAC back into .wav without any missing bits, and re-convert it into any future codec standard that may come.
I'm one of the people who think I hear degradation when going from CD to mp3. So I now only use FLAC.
FLAC is easy to use with Linux - plug it into Grip or your other favourite ripper in place of Lame or Bladeenc or Oggenc. I haven't used it with other OS'es, but I think it can be used with Windows and Mac's too.
You can read about FLAC and download it at http://flac.sourceforge.net/
-flieslikeabeagle
I would, however, like to tell everyone about a very interesting alternative codec. It's called FLAC, for Fully Lossless Audio Codec. And it is exactly that - completely lossless, meaning that there is NO lost information - nothing whatever is thrown away from the original .wav file, but through the magic of compression algorithms, compression is still possible. This is similar to the way you can zip a text file, compressing its size, but without losing a single character contained in the file when you finally unzip it.
FLAC does not produce the huge compression ratios that mp3 does - in my experience, the output .flac file is about one-third the size of the input .wav file. That means you can compress the average CD to about 200 MB instead of the original 600 MB. That still means
you can put about 5 CD's of material on one gig of hard drive space, or about one thousand CD's on a single 200 Gigabyte drive.
With the rate at which hard-drive and flash-memory capacity is increasing, in a very few years many of us will be deciding that it wasn't worth throwing away some audible quality for the sake of compressing our CD-quality audio files into mp3's. When that happens, most people will have to re-rip their entire collection. If you use FLAC, there won't be a need to do that: you can convert FLAC back into .wav without any missing bits, and re-convert it into any future codec standard that may come.
I'm one of the people who think I hear degradation when going from CD to mp3. So I now only use FLAC.
FLAC is easy to use with Linux - plug it into Grip or your other favourite ripper in place of Lame or Bladeenc or Oggenc. I haven't used it with other OS'es, but I think it can be used with Windows and Mac's too.
You can read about FLAC and download it at http://flac.sourceforge.net/
-flieslikeabeagle
- Status
- Not open for further replies.