Tested Rates:
96kbs; 112kbs; 128kbs; 160kbs; 192kbs; 256kbs; 320kbs; CD Audio

It does not exist any hardware or software allowing to judge quality of a perceptual coding scheme like the MP3. Some people use a sound signal subtraction to judge, but this method, due to the exploitation of the masking effect during the encoding process, can not be valid (it could however be valid for some comparative test between encoder parameters). It is therefore necessary to process using listening tests in order to judge quality of the encoding result.

The recording used is the 8th track from the album The book of secrets of Loreena McKennitt, distributed by Warner. It was selected due to the very good quality of this recording. The track used includes a part of choruses with some little noises in the hall, a composed part of violin accompanied by piano. A few minutes after the singer's voice is added.

My tests are done by 2 listeners (including me) at different compression rates. We both know quite well this recording, and listening are carried the weekend in beginning of afternoom in order to avoid the tiring effect that could have influenced the listening results.

The recording has been encoded with the ACM pro codec up to 128kbs, then with mpeg Encoder 0.06. The decoding was done using Winamp 1.6. The samples have then been burned on a TDK CD-R.

The listening equipment is the following :

• Teac VRDS 25 CD reader
• MIT T2 cables
• Yamaha AX 1050 amplifier
• Denon PMA 960 amplifier (for frequencies <50Hz)
• Celestion speakers

96kbs: The sound clearly lacks definition: as an example, hall's noises are perceived as some breath. The result is comparable to a good FM radio.

112kbs: The sound seems less present and less natural than the original. The definition is a bit less good, the voice is less clear. Attacks are less spontaneous. The spatialization is different from the original recording: the sound seems to be located more far and more lower. There is however a very noticeable improvement compared to 96kbs.

128kbs: Hall's noises are slightly less defined than the original. The violin is a bit less present and the piano attacks a bit less sharp. The voice is nearly identical to the original recording but sibilants are less pronounced. We can notice the same spatialization problem as with the 112kbs's one although there is again a good improvement compared to the 112kbs rate.

160kbs: The sound is more natural than 128kbs but the improvement is less spectacular than during the two preceding stages. The sound is different from the original, without however being possible to tell in what. I think that the difference resides more in what we feel rather than in what we hear.

192kbs: The sound is not felt as the original recording. It is however totally impossible to tell in what.

256kbs: The sound is indiscernible from the original. It is impossible to make the difference with the original recording.

320kbs: The sound is indiscernible from the original. It is impossible to make the difference with the original recording.

CD Audio : The sound of the burned CD is strictly identical the manufactured CD. This test, although it could appear useless, is however necessary so in order to insure that it is impossible that the burning step introduces differences, that would have falsified tests.

It is clear that the 128kbs rate does not produce a quality equal to a CD on a good quality Hi-Fi installation. We can wonder if Fraunhofer's institute has not made an error by limiting its ACM pro codec to 128kbs. However, in the context of a computer use, the quality is equal to the one obtained by reading an Audio CD on a CD-ROM reader. The quality at 128kbs is also indentical to the one obtained with the original CD on a mini or midi Hi-Fi installation, and on the vast majority of Hi-Fi installations in separated elements. The test equipment is indeed better than the majority of Hi-fi installations.

Conclusion : For a computer use, the 128kbs rate produces a quality equal to an audio CD. But in the case of an MP3 use in advanced Hi-Fi, it is necessary to use a 256kbs bitrate to reach an identical result to the CD sound.