[Vorbis] Bit rates, files size

[Shawn Riley]

I'm not sure about reconstructing a vorbis stream from raw audio based on the fact that the supposedly-inaudible information has already been removed. I think the encoder would have to determine what block lengths were used, or even if it was Vorbis at all. I think the audio would also have to be aligned at sample accuracy. That sounds too much like black magic to me. But I'm not a programming guy.
But I know this... Open up any raw sample in an application that can give you a voiceprint (spectral view) of the waveform, & if it's been lossily compressed, you'll see some very dark patches splotched all over the voiceprint (assuming that dark means quiet & light means loud). That shows you where some of the information has been removed. I see it a lot on stuff that I've recorded from TV straight into MPEG format if for some reason I have to edit the audio separately.

- Shawn Riley


At 14:13 5/10/04 +0200, Hugo van der Merwe wrote:
>Makes me curious, could one use the Vorbis model (or even the mp3 model 
>then, for that matter) to rate two samples of the same music, as an 
>indication of which might be higher quality? (With some threshold set, 
>that if the difference is too small, the software would refuse to make a 
>recommendation.) I would like somehting that could do this, or maybe 
>even just something that can determine whether a raw audio sample has 
>been lossily compressed with a certain codec (e.g. mp3 or vorbis) or 
>not. The latter should be possible, should it not? If you can measure 
>the amount of "inaudible" information in the audio, or ideally, I would 
>like Vorbis to compress to smaller sizes at -q8 if it is compressing 
>something that has been compressed and decompressed at -q3 than 
>something that is the raw original?
>
>Hmm, one too long paragraph.
>Hugo
>
>Haxe wrote:
>
>>No, there is absolutely no way to compare the audible quality of two 
>>encoders "digitally".
>>
>>Of course, you can invent some model of how the human ear might "feel" 
>>the difference, may it be square differences or anything else, but that 
>>doesn't mean that your model even remotely tells the truth about 
>>audible quality.
>>
>>Other mathematical models have been invented before. One model is used 
>>in MP3 to get the highest possible "quality" (in terms of that model) 
>>with the least possible bits. Vorbis uses another model of measuring 
>>"quality". And your sqare difference may be a third model.
>>
>>But to tell how good a certain model reflects what the ear really feels, 
>>there is NO other way than to ask lots of careful listeners in a 
>>double-blind test.
>>_______________________________________________
>>Vorbis mailing list
>>Vorbis at xiph.org
>>http://lists.xiph.org/mailman/listinfo/vorbis
>>
>>
>>  
>>
>
>_______________________________________________
>Vorbis mailing list
>Vorbis at xiph.org
>http://lists.xiph.org/mailman/listinfo/vorbis
>