Le 02/12/2011 14:15, Romain Beauxis a écrit : > Hi, > > 2011/12/2 Martin Hamant<[email protected]>: >> >> Le 01/12/2011 19:24, David Baelde a écrit : >> >>> Hi guys, >>> >>> I listened to the two files. Sometimes I feel like I can hear some >>> artifacts, but it's really slight; I'm not sure I could tell them >>> apart in a blind test. Anyway, I'm no expert in hearing compression >>> artifacts. >> I'm no expert either, but I have good hears (I don't say you haven't ;) ) >> and I can tell for sure the sound is affected, much more as it should be for >> this bitrate. >> But see below I have done more tests. >> >> >>> I'm not sure where to go from here. It would be interesting to check >>> what parameters really end up being passed to liblame with the two >>> tools. Sometimes, an interface might do some adjustment on parameters >>> to make things sound better. >> >>> Otherwise, if you don't want to dig into the source code of audacity >>> (for liquidsoap, you can ask us the details) you can just try to raise >>> a little bit the quality settings. >>> >> The thing is, even exporting with lame parameter -q9 ("Disables almost all >> algorithms including psy-model. Poor quality."), I can't get it as worse as >> the LS's one. >> I would like to find a parameter in lame that would reproduce the noise I >> can hear, but I can't find one to make it worse :D >> >> More seriously, I done some more testing. I tested a compress to 96Kbit/s >> and compared it with the 128Kbit/s from liquidsoap. The conclusion is the >> artefacts I hear in the 96k one is for sure not the kind of the >> artefacts/noise in LS one. >> the 96k bitrate raise some "birdies" in the sound when the LS file get some >> "noise"/tremolo in low frequencies. HF are not so much affected. >> >> For me, there is little chance that this quality issue to be related to >> libmp3lame. So the question would be, may LS apply some kind of processing >> on the file before or after the encoding step ? > No we don't. Actually, since you are using FLAC as input, you can > assume that PCM data fed to lame is the exact original data. > > However, I'v had a look at ocaml-lame parameters and I think this guy, > which we do not use, might be a good candidate: > http://liquidsoap.fm/modules/ocaml-lame/Lame.html#VALset_quality Like I've specified, this quality setting doesn't affect the sound in the way I have observed with liquidsoap. What I can say for sure is even q=9 (worst) from lame command-line tool is far better than current liquidsoap mp3 output thru libmp3lame :/
The more surprising in what I can hear (and see on spectrograms comparing the two resulting files) is that rumble in the end/low-end frequencies. > > I'll patch our HG source to add support for it ASAP and let you know > so you can run your tests against it. Cool ! Thanks. ------------------------------------------------------------------------------ All the data continuously generated in your IT infrastructure contains a definitive record of customers, application performance, security threats, fraudulent activity, and more. Splunk takes this data and makes sense of it. IT sense. And common sense. http://p.sf.net/sfu/splunk-novd2d _______________________________________________ Savonet-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/savonet-users
