A rather basic test was conducted using Adobe Audition and its Fraunhofer MPEG Level 3 compressor. The signal consisted of a one-second, 440-Hz sine wave sandwiched between two 0.5-second silences. The tests were made with redbook paramaters and 11:1 compression to 64 Kbps. The saved split files were then appended to provide the images below. Checking with splits and joins of WAV files showed no anomalies; repeated with lossless compression (APE) and decompression also showed flawless joins. Other compressors, both alternatives for MP3 and other formats such as WMA, can be tested similarly. I would be interested in reports on those and may include them in the primer if you permit.
The question of the gap arose from those trying to reassemble a waveform from several pieces in MP3 format. For example, they may take a concert in the form of an MP3 file for each number and burn a CD-DA in DAO. The result can be annoying clicks between tracks or even a perceived silence where the sound should be continuous. MP3 compression adds a (nominal) silence before and after each file's audio content. If there is silence or near-silence at the split, any introduced silence simply adds time and is not noticed on listening. However, if there is significant sound at the split, introducing silence has the same effect as adding a pop. The transitions to and from sound at the edges of the gap are audible. The significance varies with the loudness, the nature of the audio and the duration of the silence, but it is not unimportant for most listeners most of the time. A representative signal in MP3 is here (though not in the PDF), from Test 1.
This demonstration spilts at a null. The introduced gap is 44 msec; that is consistent across the tests and is quite audible. The transients are slight and negligible in practice, but the gap is not. The shortest found to date (other compressors, other rates) is 20 msec; the longest, nearly 100 msec.
This test splits at a peak of the 440-Hz tone. Note that the level exceeds that of the original; for that reason, compression of a maximum signal may result in clipping. Here, the transients nearly fill the gap between sound segments and have significant high-frequency components. If the segments are to be rejoined, ensuring minimum perception of the operation would require discarding a cycle of the test tone.
The results here are intermediate between those of Tests 1 and 2. Simply eliminating the gap by hand is probably sufficient for most listeners.
Adobe Audition's WMA compressor was used just as Test 1 used MP3s. While there is no gap per se, the introduced changes are evident. That is consistent with my determination that WMA generates substantial artifacts on complex audio; for that reason and despite its apparent saving of file size, it is not a format I recommend.
It is strongly recommended that track splits be at near-silence if there is a possibility that the separate MP3s will be rejoined either as files or through conversion (e.g., to CD-R). To create a CD-DA, it would make sense to have a single MP3 for the entire disc and a CUE sheet to split the tracks. While that has not been tested thoroughly, it seems to solve the problem of gaps and clicks. A convenient tool to generate the CUE sheet is CDWAV, found among the URLs at this site.
E-mail me at cdrecording@mrichter.com
Return to Mike's home page