Results
I played the MP3s
and test CD tracks through both my home stereo and my Sony MDR-V6
headphones (details listed below). The songs were encoded at a
constant bit rate, forced stereo, at 128, 160, 192, and 256 kbs. All
listening was done sighted (encoder and bit rate were known).
I first compared
the MP3s to the original CD tracks, and at 128 kbs I had no
trouble discriminating the CD from the MP3s. At higher bit rates this
became increasingly more difficult, but even at 256 kbs I believe I
could hear differences by concentrating on certain parts of the track
which were more susceptible to encoding error. However, since I did
these tests sighted, I cant guarantee that I didnt
imagine some of these differences. So please accept these
observations with a word of warning. It is very
easy to fool yourself into believing you can hear differences.
Purported differences can fade away when you do the test double blind.
"Razumosvsky"
In the
"Razumosvsky," at 128 kbs, there is a noticeable softening
of the attacks of the strings with all of the encoders. With the
Blade and LAME encoders, the viola at one point in the track produced
a "boxy" tonality that wasnt apparent on the CD or
with the other encoders. The Xing encoder sometimes sounded a bit
"whispy" on the softer, quickly-bowed sections and had a
bit of metallic tonality in general. FhG was the best of this bunch
at 128 kbs, but still not nearly CD quality.
The sound improved
a great deal at higher bit rates, as is to be expected. The boxiness
of the Blade and LAME encoders was much less prevalent at 160 kbs and
disappeared at 192 kbs. The Xing remained whispy at 160 kbs, but
sounded quite good at 192 kbs. Id be hard pressed to pass a
double blind comparison of these encoders at 192 kbs with this test
track. I was actually a bit disappointed because I thought the
Razumosvsky track would be more revealing.
"Tears in Heaven"
"Tears in
Heaven" produced a different set of artifacts. In the very
beginning of the track you can hear high frequency mike or pickup
noise. On the CD it sounds like normal, uncorrelated noise, but the
MP3s produced a swirling and swishing artifact. Also, the foot
tapping lost its eerily real presence and the bell lost its clean
decay. For lack of a better description, the Blade encoder sounded a
bit "crunchy" in the high end, which might have been caused
by an excessive brightness in the sound. The Fraunhofer encoder
produced a surprisingly harsh sounding attack on the guitar; it
remained quick and sharp, but was artificially crisp and accentuated.
Nevertheless, the Fraunhofer encoder produced the best sounding
vocals and, again, was arguably the best sounding of the bunch at 128 kbs.
Many of the
aforementioned artifacts disappeared at higher bit rates. At 160 kbs
the FhG encoder lost the harsh guitar sound. In fact, most of the
swishing and swirling high-frequency artifacts lessened at 160kbs
with all the encoders. Nevertheless, the "you are there"
imaging and presence was absent, and the tonality in the upper
register was not quite right.
I could hear even
more concrete improvement by stepping up to 192 kbs. The high
frequency tonality especially improved. For example, the percussive
bell near the start of the song decays cleanly. At 256 kbs I
dont think I could hear differences between encoders, but none
of them quite reached CD quality. It was actually the foot tapping
that gave the encoded tracks away. Listening to the CD, you can
almost see the shoe tapping the wooden stage (Im not kidding!).
Listening to the encoded tracks, you could only hear it.
"Setting Sun"
"Setting
Sun," by the Chemical Brothers, was a more revealing test track
than I initially thought it would be. In general, the sound was harsh
and unpleasant for all the encoders at 128 kbs. There's a lot of
high-frequency information in this song, and it might have given the
encoders some trouble. Specifically, all of the encoders had
difficulty reproducing one percussive track (it sounds like a
pitch-shifted snare drum). The Blade encoder was the worst offender,
adding a click or a knocking sound to this track. This was a gross
enough deviation from the sound that I would call it an encoding
error rather than just a general MP3 artifact (this went away at 160
kbs). The harshness remained at 160 kbs, but things were much cleaner
at 192 kbs. In addition to the high end harshness, the mid register
sounds became unfocused and diffuse at low bit rates. Again, this
might have been caused by problems with encoding the high frequency
overtones. The distinctive bass drum beat sounded fine at 128 kbs and above.
The envelop please:
Low bit rate
winner (128 kbs):
Fraunhofer. It was
no contest. While not CD quality, the FhG encoder was sonically the
least offensive. And with its superior measurements, it clearly beat
out the other encoders at 128 kbs. If you have a tiny hard drive or a
portable MP3 player with only 32 MB of memory, this is the encoder to use.
High bit rate
winners (192 and 256 kbs):
Fraunhofer, LAME,
Blade, and Xing (in order of preference). At the higher bit rates
they dont all sound the same, but the encoders in general were
very listenable. I found the Xing high frequency measurements a bit
worrisome, but at 256 kbs I thought the Xing encoder sounded as good
as the rest. If you are "Napsterizing"
some music, I think 192 kbs is a good compromise between sound
quality and download times. For my personal MP3 CD-R disks, I encode
at 256 kbs. This is still more than 7 CDs on one CD-R.
Parting shots
The most popular
bit rate for MP3s is still 128 kbs, by a wide margin. Certainly MP3
trading on the internet is still dominated by 128 kbs, and many MP3
web sites still encode only at 128 kbs. This is understandable. 128
kbs MP3s played over an average computer sound system is very
listenable, and listening to 128 kbs files on a portable player in an
environment with moderate ambient noise sounds just fine.
The problem with
128 kbs occurs when you take the MP3s from a typical computer desktop
environment and move them into a serious listening environment. With
the increased importance of sound reproduction for computer gaming
and DVD video viewing, the average computer speaker system is of
higher quality than it was the past. In addition, stand-alone MP3
players like the Apex AD-600 untether the MP3 music format from the
computer and bring it to the higher fidelity home stereo. Thus the
listening conditions for MP3s have changed, but the standard bit rate
of MP3 encoding hasnt. If you cant hear the difference
between 128 kbs and 192 kbs then congratulations, the more hard disk
space to you. But if you can, why settle for 128 kbs? Hard drives are
getting larger and computers are getting faster. Mpeg-1 Layer III
audio is uniquely suited to match the velocity of these technological
changes. So please, for the sake of high fidelity, lets step up
the bit rate.
Technical Info:
Computer
equipment: Celeron 450, Kenwood True-X 42x SCSI CD drive, Adaptec
2930CU, Western Digital Expert 18.1G drive, Smart and Friendly 4x
SCSI CD-R.
Ripping &
encoding: CD tracks were ripped with Exact Audio Copy v. 0.9 preBeta
3. Wav files were trimmed in Cool Edit 96. Encoders used were
described above. MP3 to wav conversion was done by Cdex 1.2. CD-Rs
were burned with Easy CD Creator 4. Programming for power spectra and
MSE measurements were done in Labview 5.1.
Audio Equipment:
Apex AD-600a, Audio Alchemy DTI (dejitter device), Audio Alchemy DDE
3.0 (D to A), AudioPrism Debut 40W tube amplifier, Reference 3a Royal
Master Monitors, Sony MDR-V6 headphones.
Copyright
2000 arstechnica.com
Next: Results
& observations
What do you
think? Post your comments on our Forums.
|