A Beginners Guide to Lossless Audio

8 thoughts on “A Beginners Guide to Lossless Audio”

1. Karl Uppiano

   January 7, 2012 at 5:28 am

   @PJ Now, lossy compression (or more accurately perceptual coding, or bit rate reduction) does reduce the number of bits to represent the audio, by throwing away data that the perceptual model indicates that you cannot hear, or would be least objectionable to not hear. The perceptual model is created using data about typical human hearing characteristics, particularly masking effects. If you hear two sounds, close to each other in pitch, the louder one will mask the softer one, making it sound as if it isn’t even there. There is no reason to encode it, because nobody can hear it. The model keeps track of how far apart in pitch and loudness the two sounds have to be before the masking effect stops working (in which case both sounds do have to be encoded). The model also leaves a safety margin for differences in human perception, and tends to encode for the worst case, although at very low bit rates, the margin might be insufficient, and people will hear the loss in quality. That’s the price you pay for very low bit rates.

   The compression algorithm analyzes sounds in the entire audio spectrum, and allocates bits from the target bit rate and attempts to encode the audio with the least objectionable side-effects. The higher the target bit rate, the less likely there will be audible artifacts.

   Single generation compression at reasonable bit rates can sound very good, but multiple generations, and re-mixing using compressed and reconstituted audio might have noticeable deterioration, rather like making copies of a copy of a cassette tape, which was the analog equivalent of 64Kbps MP3.
2. Karl Uppiano

   January 7, 2012 at 5:09 am

   @PJ, there’s a very easy way to determine if a compression format is lossy or not. Get a copy of Audacity, and import the original sound. Use the invert effect to invert the track (which flips the polarity, so a 1 becomes a -1, etc.). If you then import the original track again and play them both in exact time alignment (which is easy to do in Audacity because you can slide the tracks relative to one another). At maximum zoom, you can align the samples exactly. With zero loss, the tracks should sum to zero — absolute silence, meaning the difference between the reference copy (the original sound) and the test copy is zero. If there is any difference at all, you’ll hear it (as noise, distortion, or whatever). That’s the control test. Absolute silence means an exact copy. Any residual signal at all indicates loss or distortion.

   I told you that so I can tell you this: Lossless audio compression is not bogus. It is very real. It just isn’t as effective as other forms of bit rate reduction, but it doesn’t have the disadvantage of distorting or deteriorating the original sound. Anyone can prove that for themselves using the technique I described above.

   Here’s how lossless compression works: In most stereo recordings, there is a lot of redundancy between the left and right channels. A simple mathematical calculation can store left + right as a single channel (using half the space), and simultaneously store the difference between the two channels (left – right) using fewer bits. This can reduce the storage requirement from by up to 1/2 the original size. There are many other features of sampled audio that can benefit from mathematical analysis, resulting in smaller file size, and which are completely reversible.

   Don’t be fooled by listening tests. The ear is notoriously unreliable for comparisons except in very well matched, double blind A/B tests.
3. byerh

   March 14, 2011 at 12:54 pm

   PJ AudioSnob With all due respect, I do not beleive that you are correct: My understanding is that lossless compression is like zip compression but refined specifically for audio. Your analogy would say that when zipping a word document zip mathematician/programmers leave out the letters and/or words that are not important and do not 100% restore the document to its original state when unzipping. Since we all know that is not the case why do you beleive that lossless cannot be restored 100%. It can – that is what it does. See http://flac.sourceforge.net/
4. mark morrison

   January 13, 2011 at 7:15 pm

   PJ – excellent work
5. PJ AudioSnob

   December 29, 2010 at 12:00 pm

   @ everyone considering lossless

   I honestly am disturbed about this whole lossless debate. Ultimately, what compression does is remove certain data based on the “educated guess” by the mathematician / programmers that their formula can replace that audio information EXACTLY as it was when decompressing it.

   I’m no genius, but decompression is just running the formula on the compressed data. The compressed data must contain “clues” so that the decompression formula knows where to add things back in. Hence the CPU hit of doing this in real-time.

   Now, since that is my understanding of “how it works”, from a layman-techy’s perspective, allow me to say the following:

   WHY I THINK LOSSLESS IS A JOKE:
   First, the above. They are “guessing”. The only way I’d believe them is if they showed me a high sample-rate side-by-side comparison of lossless vs direct CD (or even DVD-Audio) which showed all the “waves”. If they don’t match 100% at ALL frequencies (even those considered inaudible by many math-heads), then I would have to conclude that loss has occurred. For this to be a valid comparison, I imagine the sample-rate would have to be very high. No cheating by claiming I couldn’t hear the difference (and I’ll tell you why in a second).

   The MAIN reason I think lossless is a joke, however, is not because of my own educated guessing. It’s because I can flat-out HEAR the difference. I’ve complained about this since MP3 first came out. Audio-compression BUTCHERS acoustic instruments, typically in the high and low frequencies. For a comparison, go find a CD with a high-quality studio recording of any music where you can clearly hear the following:

   1. Cymbals / Hi-Hats / Snares
   2. Acoustic Guitar (nylon and steel strings)
   3. Piano
   4. Violin / Cello
   4. Any brass instrument

   Now, I challenge you to run a track through FLAC with all settings at the highest possible quality. This will result in a file size that is 90% as large as the original file. Should be lossless right? Listen to both tracks, back and forth. Listen closely to the details in the instrument(s) I listed above. If your ears and brain are like mine, and you are picky about audio quality, I am willing to bet you will conclude that the compressed file sounds a little “dead” or “flat”. Certain frequencies are just plain MISSING.

   Also, listen closely to lower frequencies (Cello, Bass strings, certain ranges of Piano notes). Doesn’t the compressed file make the bass sound a little muddier, less defined, maybe a bit boomier. To me it does. I’d be curious to hear your feedback.

   **** The bass test was done on my home system, a Denon home theatre amp (~$500) and 2 Kenwood speaker cabinets (~$350). Audio input source was my HP laptop going out the headphone jack to RCA plugs into my amp.

   **** I have also done this test with a Dell Optiplex and a cheap set of desktop Dell Speakers, and could tell the difference, though it was a little less obvious.

   But seriously, for Jazz, Classical, and any type of music that relies heavily on accoustical instruments, lossless is for the birds. I’ll be upgrading my mobile music player to the highest possible data capacity to be able to listen to my audio as close to original as possible.
6. Sven

   June 30, 2009 at 9:41 am

   @MIKE …u kidding me? Take a PCM file and run it with ZIP or RAR arciver, then decompress… any difference? NO period. Same with FLAC or any other compression that is lossless. It’s a mathematical process and NOT a lossy way that inwolves throwing away parts. Read and learn before giving your opinion.
7. David Jacques

   April 27, 2009 at 4:37 pm

   When converting flac to wav, what is the difference of converting at 44.100 khz 16 bit stereo 172 kb/sec and 48.000 khz 16 bit stereo 188 kb/sec? Which gives the best quality? Do all conversion methods of flac to mps result in the same quality? I am using winamp to wav and then itunes for wav to mp3. This method is free but I’m curious if some of the software that can be purchased results in a better (cleaner)mp3 result.
8. Jason

   May 10, 2007 at 12:03 pm

   nvm guys i got it