I did a small project using various lossy encoders to see how they all encoded digital silence with their VBR methods, and here are the results. The bit rates are reported as they appear in Foobar2000. The encoders and settings used are written in the readme.txt file.

Click to view attachment1. Ogg Vorbis (1 kbps)
2. AAC (2 kbps)
3. MPC (3 kbps)
4. MP3 (32 kbps)
5. WMA (53 kbps)

Its important to know, that those results mostly reflect the limitations of the various codecs themselves. For example, the minimum bitrate for mp3 is always 32kbit - it cannot go below that. Vorbis instead scales its bitrate seamlessly. Anyways, with normal music, this is rarely an issue, because digital silence almost never happens.

- Lyx

I suppose this isn't much of an issue overall since there is not likely much digital silence in an average song. That said I am still interested in the reasons why certain encoders store it as they do. I was aware that for higher bitrate presets in LAME, the bitrate "floor" was 128kbps, with the sole exception being 32kbps for digital silence. I did not however know that vorbis would go as low as 1kbps for it!

Again, not that it seems to have much relevance in the long run, but I'm just curious as to why if the sample is digital silence, 32kbps (or more for WMA) are required? It seems strange to me. Could this be one reason why vorbis files yield, on average, higher overal perceived quality at the same bitrates?

LAME has a high bitrate floor, but it makes an exception for silence.

The WMA encoder probably doesn't have that exception.

With mp3, it is a limitation of the format itself. With WMA, i dont know.


No. You said yourselves that digital silence is rare.

Great work.

Although silence is not much of an issue in practical manners, this somehow shows the codecs limitations, like Lyx stated before. And, in this case we can clearly see how MP3 in conjunction with WMA are such limited formats. Even more incredible fact is to see how popular both still stand today.

Well, the OP should have had a look at what Windows Media Player would have reported about the bitrate of WMA instead. Foobar tends to display wrong average bitrates of VBR WMA files, as I have observed in numerous occasions so far. In every case the actual bitrates were much lower compared to what foobar had claimed in the properties window.

Here's what WMP reports to me about your wma file:
Bit rate: 51 Kbps
Audio codec: Windows Media Audio 9.2, VBR Quality 75, 44 kHz, stereo 1-pass VBR

I encoder the silence.wav file using Winamp 5.35 with WMA 9.1 16 bits, stereo, 44 kHhz VBR Quality 10 and got a file with 25 Kbps. Finally proof that that WMA can produce CD quality digital silence at a lower bitrate than MP3.

Lossless codecs compress silence at less than 1 kbit/s

The silence encoded by FLAC 1.1.4 in Easy CD-Da extractor has 1 kbit/s bitrate while the same file encoded by command line version of FLAC 1.1.4 has 2 kbit/s.

Maybe it's due to padding or different number of seek points.

How about an abx test on silence. Also, I assume only the highest of hi-fi equipment can accurately play multi-channel surround silence, with all its nuance and depth and feeling.

I ran your silence.mp3 through SoundSlimmer and got an amazing 582 byte file. That's a compression rate of 0.24% (or 0.0097kbps) of the mp3 file. Decodes faster than real time on my P4 2Ghz. It appears MPZ can be good when encoding silence.

I don't listen to much 'mainstream' music, but I was under the impression that most commercial CD's have silence at both the beginning (usually around 1/3->1/4 of a second) and end (2-3, sometimes 3-5 seconds) of a track.

That would become an issue with MP3, if it encodes 5 seconds of silence per track at 31 kbps higher than Vorbis.

Anyway- my point being, people are thinking of silence in the actual musical track, but it's far more likely to occur at the start/end of the track.

Unless everyone here truncates silence

- Spike

I use Vorbis for encoding my cds, and i have found the low bitrate for silence usefull for cds with hidden tracks. The most extreme was about 4 minutes of music followed by maybe 15 minutes of silence and then 3 more minutes of music, so the file size was equal to about 7 minutes of music instead of 22 minutes, so in this instance it was nice. However, i then split the file into two files, ignoring the silence, so in the end it didn't save me anything

In my opinion the important thing is not if encoding digital silence or not, that's more like trivial. What worries me (and might also worry you) is the fact that MP3 codecs do not encode in real VBR, which means it will always use one of the pre-defined and limited numbers (32, 40, 48, 56, 64, 80, 96, 112, 128, 160, 192, 224, 256 or 320kbit) for each audio sample, which could either result in a bigger filesize or even degrade in quality (this is where I think the developers should work their brains out and work each sample with some quality headroom just to make sure quality is ensured for every sample).

All in all, this test was useful though since it has shown us the various lossy formats VBR limitations. In this case, Musepack (.mpc) and Ogg Vorbis (.ogg) are champs.

I think you mean frame, and in any case, I believe you're wrong.

Yes, mp3 frames are fixed at those values (they're about 10% increments), but the bit reservoir means you can effectively push the bits around between frames. AFAIK this means the only limits on VBR are 32kbps and 320kbps. For everything in between, you use what you want, and add to / subtract from the bit reservoir to reach one of the allowed frame sizes.

The bit reservoir also means that silence at the start and end of songs isn't quite as big a deal as it seems.

http://www.hydrogenaudio.org/forums/index....showtopic=38444

Cheers,
David.