Case Study mp3 - Detective Needed

Case Study mp3 Soundtest / Home

Detective Needed

The hissing and superimposed distortions appear all too typical for mp3. For comparison, another 128kb/s "Without You" was downloaded from the Internet. The differences were very apparent. Less hiss, but a heavier tone and the pulses at the beginning gave an impression of spatial instability. Some listeners found this recording more pleasant. However, the differences could originate from the original, which has a noticeabley higher tone on our CD "Music Box". A glance at the frequency spectrum made everything clear; our original and the mp3 version showed good correlation whereas the Internet version showed obvious attenuation of frequencies above ca. 15kHz. An additional check was performed by increasing the bitrate to 256kb/s. The compression is then 5.5fold and the tone quality still lies below the middle between the CD original and the 128kb/s version, but still nowhere close to the analog recording. This holds for the other encoders as well, although the LAME encoder continually comes very close to the analog recording. The decoder is considerably less critical and also requires processing time. The check using the Mahler recording with 128kb/s and 256kb/s brought similar results, although the mp3 distortion was less conspicuous and only minimal for the higher bitrate. The cause is probably that despite the large dynamic range of orchestra music, the overlay effects end up making the compression effects practically imperceptible. For none of the recordings was the mp3 version considered more pleasant or at all better than the CD.

Finally, reduction by factors 8 and 32 and subsequent multiplication to the original level was used to manipulate the recording to correspond to resolution of 13 and 11 bits, respectively, rather than the original 16 bits. Even these low-fidelity versions still sounded better than mp3, although the 11-bit version had a very apparent noise similar to analog. This does not mean that 13 bits would suffice for the CD. After all, the rest of the signal path is on 16-bit resolution. The savings on memory requirements attained through resolution reduction is very low compared to that achieved using mp3 compression. Loss-free compression techniques in audio technology are not particularly effective (approximately 30% improvement, depending on the music) compared to graphics, for example. It should be noted that a number of the CDs (e.g. the Mahler recording) remained under the maximum level at even the loudest point and thus unfortunately do not take full advantage of the resolution.

An interesting point is that the mp3 recordings contained tone distortion effects which were unlike any previously known from analog equipment and CDs. These occur primarily when the human hearing deception effects used by mp3 could not be taken advantage of.

The results of the loudspeaker tests were greatly anticipated. Noise and small disturbances are less apparent in this case. This was verified in practice. Slight discrepancies could be perceived, but it was more the determination that they were present. In the headphone tests, it was possible to actually clearly describe the effects. An example of this was the slight overdrive of the Beethoven recording. The differences heard over average speakers were smallest for the speech and Tom Jones recordings. The playback volume used in the tests plays a major role. The louder it was, the easier the disturbances could be detected and described. It was not always simple to separate the mp3 distortions from tones in the original, e.g. as in the Fourth Symphony. In which the brass does not always sound clean. The Tom Jones recording had similar conditions.

The next stage was the speaker blindfold test. There were two rounds: The first in which the type of processing was known, and a second which was a true blindfold test.

The results make clear that one does not require a high-end system to detect mp3 distortion, an upper midrange system is sufficient. The CD player is relatively unimportant; the mp3 artifacts can be found without any problem with a PC using the soundcard and headphones. After all this DDD and HiFi, a test was made to see just how bad a recording can be so that mp3 is certain not to degrade it. The first minute of Dusty Springfield’s 1963 hit "I Only Wanna Be With You" was investigated. In this case, the recording was marred by a slight loss of high tones and knocking. The mp3 versions all contained hissing, which was stronger or weaker depending on the encoder. The NexEncoder had irritating background noise. Mp3 did not produce CD-like quality in this case, either, although the non-hifi listener may be satisfied. Similar poor tone quality recordings downloaded from the Internet (e.g. "Block Buster" from the Sweet) produced similar results. In the best cases, the impulses were somewhat eroded and the dynamic reduced. There were also songs which contained numerous encoding errors in which the encoder was obviously inadequate. Other standards such as CD or compact cassette do not truly guarantee a minimum level or even a uniform tone level. These tests made it clear that the mp3 software solution fails to do this as well.

Feedback

After the hearing test a comparison to the more technical tests is interesting. In the case of mp3 we can state that the correlation is really good. LAME and FHG are best, and there is a quite big gap to the others.

Conclusion

To Top - To Homepage - German Version