- Feb 8, 2001
- 8,968
- 16
- 81
PLEASE READ ALL THE WAY UP TO "METHODOLOGY" BEFORE POSTING
OK, so here it is. Lossy vs. Lossless can you hear the difference?
The way things work is as follows: I have uploaded several files. The first file that you must listen to is the Master Copy. This is your reference, it is lossless encoding directly from CD to FLAC.
There are then 4 files that each are compressed using the following different schemes:
A) Lossless FLAC
B) AAC CBR 320 kbps, then transcoded to lossless FLAC
C) MP3 CBR 320 kbps, then transcoded to lossless FLAC
D) ??? surprise compressor, then transcoded to lossless FLAC. Not necessarily the same compressor for each track, may or may not be lossless any of the tracks.
The file names have been mixed. You tell me which one sounds identical to the master by voting in the poll, and preferably also by adding a post. The poll is for statistical purposes whereas the post is for your own personal satisfaction.You are only required to identify the lossless setting, but you can choose to rank the files by quality should you desire to do so in your post.
Playing the Files
To listen to these files, I recommend using foobar2000 normal edition, available here foobar2000 link
FLAC installer (includes Winamp plugin)
Important Notes:
- The Master file has a different size because a different FLAC compression setting was used. The sound is absolutely identical to the lossless clip that you must identify.
- If the surprise compressor turns out to be lossless, I'll accept either of the lossless choices as being correct.
- A larger file size doesn't imply a better source, so don't base your decision on file size.
- The answer key will be posted sometime tomorrow, June 2nd, so don't nag me about answers right away. I won't send them by PM.
Audio Tracks:
Notorious B.I.G. - Hypnotize
MASTER
Candidate 1
Candidate 2
Candidate 3
Candidate 4
Billy Talent - Try Honesty
MASTER
Candidate 1
Candidate 2
Candidate 3
Candidate 4
Wagner - Ride of the Valkyries
MASTER
Candidate 1
Candidate 2
Candidate 3
Candidate 4
Peter Gabriel - Passions
MASTER
Candidate 1
Candidate 2
Candidate 3
Candidate 4
Methodology:
1. Using EAC secure mode, I have ripped one track of three CDs PCM wav files
2. File is loaded into Nero Soundtrax and cut to 46 seconds.
3. Nero Soundtrax is used to export a 46 second PCM wave file
4. Nero Soundtrax is used to export a FLAC file with compression set to 4, this is the "reference" file.
5. Nero Soundtrax is used to export a FLAC file with compression set to 8, this is the lossless file.
6. Nero Soundtrax is used to export an AAC file, set to use the Low Complexity algorithm with 320kbps CBR
7. The 46 second PCM file is encoded using LAME using --alt-preset insane --noreplaygain
8. The 46 second PCM file is encoded using the mystery encoder
9. The AAC lossy file is opened into Nero Soundtrax and exported to FLAC using compression set to 8.
10. Repeat step 9 for the MP3 file
11. Repeat step 10 for the mystery encoded file
12. Names changed to generic names and files uploaded to server
Notes about the results
I ran some quick probablilities to give you an idea of predicted results based on random numbers. If no one can tell a difference and they are just guessing:
- There's approximately 68% chance that they get at least one correct response out of four
- There's approximately 42% chance that they have exactly one answer right
- There's approximately 21% chance that they have exactly two answers right
- There's approximately 4.7% chance that they have exactly three answers right
- There's approximately 0.4% chance that they guess all four correctly
So anyone getting 3 or 4 right is almost certainly telling the truth, getting two right is a little iffy, getting one or zero right means they're probably guessing.
How big is the difference numerically?
I had a little time to kill so I ran the original and the worst compressor through a little matlab code. The mean SNR of aprox. 35 dB, so the compression noise is on average about 60 times smaller than the music itself. Obviously this is a mean SNR so there are portions where the difference is much more significant.
The maximum difference between the compressed and the uncompressed is about 10% as big as the peak signal level, so in some areas it is very signficant. Realistically though, this large difference is probably in areas where the codec is taking masking effects into the account and you probably won't hear it anyway.
I also compressed tested the same track using LAME and --alt-preset standard. This yields an SNR of 26.7dB, which is still pretty large.
Here are the SNR graphs, for those interested:
The original track is in blue, the compressed track is in red and the green is the absolute difference between the two.
LAME @ insane
LAME@standard
Answer Key
I used the same mystery encoder for all the tracks, it was
LAME --alt-preset insane -k, the switch used to remove the low pass filter that is usually applied by LAME. For convenience's sake the results below refer to this encoder as MP3-2
Notorious B.I.G. - Hypnotize
1 - MP3-2
2 - lossless
3 - AAC
4 - MP3
Billy Talent - Try Honesty
1 - lossless
2 - AAC
3 - MP3-2
4 - MP3
Wagner - Ride of the Valkyries
1 - AAC
2 - MP3
3 - MP3-2
4 - lossless
Peter Gabriel - Passions
1 - MP3
2 - AAC
3 - lossless
4 - MP3-2
Stay tuned for my next shootout which will be to determine what bitrate you need for AAC transparency for a few clips. Obviously, this is dependant each system, but it should be interesting to see the results.
Thanks for participating
OK, so here it is. Lossy vs. Lossless can you hear the difference?
The way things work is as follows: I have uploaded several files. The first file that you must listen to is the Master Copy. This is your reference, it is lossless encoding directly from CD to FLAC.
There are then 4 files that each are compressed using the following different schemes:
A) Lossless FLAC
B) AAC CBR 320 kbps, then transcoded to lossless FLAC
C) MP3 CBR 320 kbps, then transcoded to lossless FLAC
D) ??? surprise compressor, then transcoded to lossless FLAC. Not necessarily the same compressor for each track, may or may not be lossless any of the tracks.
The file names have been mixed. You tell me which one sounds identical to the master by voting in the poll, and preferably also by adding a post. The poll is for statistical purposes whereas the post is for your own personal satisfaction.You are only required to identify the lossless setting, but you can choose to rank the files by quality should you desire to do so in your post.
Playing the Files
To listen to these files, I recommend using foobar2000 normal edition, available here foobar2000 link
FLAC installer (includes Winamp plugin)
Important Notes:
- The Master file has a different size because a different FLAC compression setting was used. The sound is absolutely identical to the lossless clip that you must identify.
- If the surprise compressor turns out to be lossless, I'll accept either of the lossless choices as being correct.
- A larger file size doesn't imply a better source, so don't base your decision on file size.
- The answer key will be posted sometime tomorrow, June 2nd, so don't nag me about answers right away. I won't send them by PM.
Audio Tracks:
Notorious B.I.G. - Hypnotize
MASTER
Candidate 1
Candidate 2
Candidate 3
Candidate 4
Billy Talent - Try Honesty
MASTER
Candidate 1
Candidate 2
Candidate 3
Candidate 4
Wagner - Ride of the Valkyries
MASTER
Candidate 1
Candidate 2
Candidate 3
Candidate 4
Peter Gabriel - Passions
MASTER
Candidate 1
Candidate 2
Candidate 3
Candidate 4
Methodology:
1. Using EAC secure mode, I have ripped one track of three CDs PCM wav files
2. File is loaded into Nero Soundtrax and cut to 46 seconds.
3. Nero Soundtrax is used to export a 46 second PCM wave file
4. Nero Soundtrax is used to export a FLAC file with compression set to 4, this is the "reference" file.
5. Nero Soundtrax is used to export a FLAC file with compression set to 8, this is the lossless file.
6. Nero Soundtrax is used to export an AAC file, set to use the Low Complexity algorithm with 320kbps CBR
7. The 46 second PCM file is encoded using LAME using --alt-preset insane --noreplaygain
8. The 46 second PCM file is encoded using the mystery encoder
9. The AAC lossy file is opened into Nero Soundtrax and exported to FLAC using compression set to 8.
10. Repeat step 9 for the MP3 file
11. Repeat step 10 for the mystery encoded file
12. Names changed to generic names and files uploaded to server
Notes about the results
I ran some quick probablilities to give you an idea of predicted results based on random numbers. If no one can tell a difference and they are just guessing:
- There's approximately 68% chance that they get at least one correct response out of four
- There's approximately 42% chance that they have exactly one answer right
- There's approximately 21% chance that they have exactly two answers right
- There's approximately 4.7% chance that they have exactly three answers right
- There's approximately 0.4% chance that they guess all four correctly
So anyone getting 3 or 4 right is almost certainly telling the truth, getting two right is a little iffy, getting one or zero right means they're probably guessing.
How big is the difference numerically?
I had a little time to kill so I ran the original and the worst compressor through a little matlab code. The mean SNR of aprox. 35 dB, so the compression noise is on average about 60 times smaller than the music itself. Obviously this is a mean SNR so there are portions where the difference is much more significant.
The maximum difference between the compressed and the uncompressed is about 10% as big as the peak signal level, so in some areas it is very signficant. Realistically though, this large difference is probably in areas where the codec is taking masking effects into the account and you probably won't hear it anyway.
I also compressed tested the same track using LAME and --alt-preset standard. This yields an SNR of 26.7dB, which is still pretty large.
Here are the SNR graphs, for those interested:
The original track is in blue, the compressed track is in red and the green is the absolute difference between the two.
LAME @ insane
LAME@standard
Answer Key
I used the same mystery encoder for all the tracks, it was
LAME --alt-preset insane -k, the switch used to remove the low pass filter that is usually applied by LAME. For convenience's sake the results below refer to this encoder as MP3-2
Notorious B.I.G. - Hypnotize
1 - MP3-2
2 - lossless
3 - AAC
4 - MP3
Billy Talent - Try Honesty
1 - lossless
2 - AAC
3 - MP3-2
4 - MP3
Wagner - Ride of the Valkyries
1 - AAC
2 - MP3
3 - MP3-2
4 - lossless
Peter Gabriel - Passions
1 - MP3
2 - AAC
3 - lossless
4 - MP3-2
Stay tuned for my next shootout which will be to determine what bitrate you need for AAC transparency for a few clips. Obviously, this is dependant each system, but it should be interesting to see the results.
Thanks for participating