Recently I bought a harddrive mp3 player | recorder which I took as an opportunity to look at current audio compressors using the psycho-acoustically optimized mpeg 1 layer 3 audio technology. As space is kind of an issue (even with 6 gigs of storage *G*),
the best possible compromise regarding encoding speed, size on disk and overall audio quality - not necessarily in that order - was to be found.
Earlier (bad) experience and a short read-up on recent versions disqualified Xing's encoder that still doesn't seem to be up to standard. From the remaining few I chose lame [www.mp3dev.org] because of the
continous development and refinement of its algorithm by the openSource community and set it up with Fraunhofer's current reference implementation (FastEnc, as included with the MusicMath Jukebox). To be able to measure the technical
advances since the 'earlier days' I included mp3enc.
As I was looking for optimal results with my mobile player I used different compressions settings, focussing on VBR (variable bitrate encoding). Why? Because I think VBR's a good idea .)
You will find graphs of the results for every encoder and settings on individual pages. As I didn't want to compress the images these can take some time to load. For a quick review of my findings please refer to the
Audio equipment used for listening tests:
- Archos Jukebox Recorder [www.archos.com]
- Creative Soundblaster live! 5.1 platinum (analysis of jukebox signal)
- Yamaha preAmp / PA (MX 70)
- Sonofer speakers
- Sony Audio CD player
- beyerdynamic and AKG headphones
For realistic conditions I tried to choose high quality recordings of 'typical' material. As we'll see I had to discover that
not everything sold even as reference material for audio tests is really decently recorded material. The following pieces served as
- "Walking on the Moon" - The Yuri Honing Trio. Original by Sting. [Hörtest-CD, stereoplay].
Jazzy style brass and sax.
- "Scherzo (Allegro Vivace)" - Ludwig van Beethoven, Symphony no. 3. [Hörtest-CD, stereoplay].
Classic stuff, big orchestra - a real challenge. Unfortunately I found some flaws in the recording. For one there's a (even audible) hiss on the
right channel - and in very loud parts there seem to be errors due to exceeding the 0 dB limit in the (digital, DDD) recording. I was disappointed ...
Below is a cropped image with enhanced contrast to show the 'line' at 16 kHz in the spectrum.
- "Kassensturz" - Sigi Schwab and Percussion Academy. [stereoplay highlights collection].
Very nice guitar work by a master of his instrument - and a very good recording, showing the enormous dynamical range of digital recordings. Makes the vinyl version of the same song sound really faint and unspectacular :)
- "Toccata" - sky. [stereoplay highlights collection].
A good example of what I like to call 'bombast rock' and a re-vamp of the classical piece. Normally not exactly my taste .)
Please don't take this as an indicator to my favourites in music :) I just thought these might do good as examples.
Before anyone asks: for all the obvious reasons I'm not going to put up the original data for downloading .)
Things I noticed when analyzing the digitally extracted audio data:
- The first two pieces are full range recordings
- The rest are filtered to just about reach 20 kHz. In fact most of today's recordings are - certainly most rock or pop recordings, as this is a general recommendation in audio engineering.
It kind of makes sense: as one approaches the Nyquist freqency (about 22 kHz) of the data, moirée may become an audible problem. Most grownups won't actually hear more than - say 17 kHz at best.
I myself don't do much better - my hearing ends at about 18 to 19 kHz at medium sound levels.
That's also the reason why I very much agree with the recommendations of www.r3mix.net thus setting the compression preferences for both encoders to the recommended settings mentioned there.