Recording "lossy" formats to other "lossy" formats

[juncers]

I'm trying to understand compression algorithms and haven't found any info on the following question ( probably because it makes no sense in the real world).

What happens when you record a lossy format (mp3 @ 320k bps) to another lossy format ( minidisc @ 292k bps) or the reverse. Is the result similar to the source or do two different algorithms mixing together create something completely different? I assume the latter.

I know someone is going to say "try it and find out" - I have. I'm more interested in whether the algorithms automatically compress the source by 4X or 5X, or recognize the lossy source and compress it by only 9% (in the case of mp3 to minidisc), or something in between. I know the result will "read" at its programmed output (320 or 292), but that doesn't tell me how it got there. Just curious.

Thanks in advance,
Ray

 

[SaSi]

To start with, it depends on the flow of audio from one format to another.
For example, if you have recorded a track in 320kbps MP3 and you have that on your computer to play that back and send it via the sound card to a minidisk deck to record at that proprietary format, the data will be used to convert the stored information (MP3) into an analog signal (as best as the codec used can do) and will probably have some slight degradation from the analog electronics of the sound card. The signal will then be received in analog from the minidisk and will be digitized once more using the built-in hardware codec of the deck into the proprietary format. So you can see how "generational loss" can stack up.

However, there is something that may be mis-interpreted here.
The original lossy digital recording did forever discard some parts of the signal the codec design considers redundant. The idea is that upon playback, the resulting waveform will be almost as good as the original one - to your ears and mine - but the playback waveform will not be identical to the original one. Whatever limitations are built into the playback amplifier of the sound card or the input amplifier/filter of the receiving deck, will be applied to that waveform (let's call it a simpler version of the original waveform) and will slightly modify it, adding whatever kind of distortion it will add (harmonic, IM, noise, etc).
This means that the second codec used to digitize the signal will encode a third version of the original waveform, let's call that "distorted simpler waveform".
Since most of the lossy codecs make similar assumptions and choices on what to discard, the second encoding will find that most of what would be discarded from the original waveform is already discarded but some additional artifacts have been added to the "distorted simpler waveform" it works on.

So, each generation of decode/encode doesn't really remove (say) 30% of the content but perhaps removes the subtle artifacts added as distortion.
On the other hand, re-encoding the MP3 track, even by software, to avoid any analog electronics distortion, to a different codec (say from MP3 to AAC) means that the simplified waveform recreated from the MP3 data are re-encoded using a different strategy which may result in some different parts of the waveform to be lost.

So as to answer your question, the algorithms don't really know and identify how the original track was encoded because, even in s/w, the MP3 track will be decoded to discrete digital data (say WAV) and then re-encoded by another codec. There is no such thing as MP3/AAC or MP3/ATRAC conversion process.

And the percentage of reduction needs not be a constant. When you encode into MP3 (and actually in almost any codecs, including MPEG2/4 for video, you can tell the encoder to either aim for a constant average bitrate (say 320kbps for MP3, 8MBPS for video) or aim for a constant "quality factor". It's better to aim for the second target as it allows a lower bitrate for parts of the content that don't need that allowing an increased bitrate for parts of higher complexity without degrading the quality of parts with a lot of transients.

 

[awillia6]

Wonderfully clear and nuanced explanation. Thanks!

 

[juncers]

Yes indeed--thank you for that succinct explanation. Much appreciated.

Best,
Ray