I just made a recording of the main lead from a song i'm remaking on my Commodore 64, it has led to some strange results.
The original file in flac
--vbr-new -V2 (3.97 beta 2)
Fraunhofer VBR at 70% quality (Fraunhofer codec found in Adobe Audition)
The lame codec adds a lot of noise to this file for some reason, any ideas? It's also audible at other bitrates.
Full Version: --vbr-new -v2 vs Fraunhofer
Indeed, this is quite.. disturbing. It sounds like shit. Did you try without --vbr-new, just to check? I'm currently using 3.98a3 and it's not as noteable, but still noteworthy (abx8/8 in both cases, anyways)
I tried it without --vbr-new and also with just normal cbr encoding. At higher bitrates (160kbps++) the noise is barely audible (gone at 320 I think), but as you decrease from 160 it gets worse and worse.
QUOTE(locu @ May 2 2006, 09:28 PM) 
I tried it without --vbr-new and also with just normal cbr encoding. At higher bitrates (160kbps++) the noise is barely audible (gone at 320 I think), but as you decrease from 160 it gets worse and worse.
I guess this is a major problem sample, then.. Better notify Gabriel of this.
Its not surprising given that such artificial samples are know to cause problems with all transform coders.
QUOTE(shadowking @ May 2 2006, 09:50 PM)
Its not surprising given that such artificial samples are know to cause problems with all transform coders.
And FhG doesn't use MDCT?
No noticeable noise at 96kbps AAC VBR in iTunes
This is a very interesting sample.
It shares some similarity with two of the worst samples I know, trumpet and herding_calls, in that a very audible additional signal is added. In this case it sounds like noise, in case of trumpet and herding_calls it's rather like an added distorted signal.
This sample is pure mono which takes only 114 kbps on average for the FhG encoding (going up to 192 kbps at maximum). So it is not a hard problem for the mp3 format.
I know from my other very bad problem samples that 3.98a3 behaves much better with severe problems, but in this case even 3.98a3 -V0 was not too bad at best.
My hypothesis has always been that 3.90.3 when avoiding VBR behaves much better than 3.97/3.98 for very severe problems, and as for them ABR 224 kbps is sufficient to bring them to my not-at-all-annoying level (no matter whether GPSYCHO --abr 224 -h or NSPSYTUNE --alt-preset 224), and ABR 256 kbps brings them near to perfection. Same bitrate CBR BTW does the job as well.
I tried, and it proved true also for this sample.
As this sample is pure mono I also tried low bitrate out of curiosity and challenged by FhG. 3.90.3 --abr 128 -h is quite allright, however --alt-preset 128 isn't quite so (but still a lot better than the 3.97 -v2 encoding).
Quite interesting as it confirms what I found (to a much lower degree) with 3.90.3 GPSYCHO and 3.90.3 NSPSYTUNE in ABR 192 kbps and ABR 224 kbps mode (whereas at 256 kbps the differences are gone for me).
So roughly speaking for these problem samples to me qualitywise at the moment it really looks like
3.90.3 GPSYCHO > 3.90.3 NSPSYTUNE > 3.98a3 NSPSYTUNE > 3.97b2.
For very good quality in an overall sense my personal favorite is 3.90.3 NSPSYTUNE as for now: --alt-preset 270.
It shares some similarity with two of the worst samples I know, trumpet and herding_calls, in that a very audible additional signal is added. In this case it sounds like noise, in case of trumpet and herding_calls it's rather like an added distorted signal.
This sample is pure mono which takes only 114 kbps on average for the FhG encoding (going up to 192 kbps at maximum). So it is not a hard problem for the mp3 format.
I know from my other very bad problem samples that 3.98a3 behaves much better with severe problems, but in this case even 3.98a3 -V0 was not too bad at best.
My hypothesis has always been that 3.90.3 when avoiding VBR behaves much better than 3.97/3.98 for very severe problems, and as for them ABR 224 kbps is sufficient to bring them to my not-at-all-annoying level (no matter whether GPSYCHO --abr 224 -h or NSPSYTUNE --alt-preset 224), and ABR 256 kbps brings them near to perfection. Same bitrate CBR BTW does the job as well.
I tried, and it proved true also for this sample.
As this sample is pure mono I also tried low bitrate out of curiosity and challenged by FhG. 3.90.3 --abr 128 -h is quite allright, however --alt-preset 128 isn't quite so (but still a lot better than the 3.97 -v2 encoding).
Quite interesting as it confirms what I found (to a much lower degree) with 3.90.3 GPSYCHO and 3.90.3 NSPSYTUNE in ABR 192 kbps and ABR 224 kbps mode (whereas at 256 kbps the differences are gone for me).
So roughly speaking for these problem samples to me qualitywise at the moment it really looks like
3.90.3 GPSYCHO > 3.90.3 NSPSYTUNE > 3.98a3 NSPSYTUNE > 3.97b2.
For very good quality in an overall sense my personal favorite is 3.90.3 NSPSYTUNE as for now: --alt-preset 270.
Should be solved in 3.98a4 (just commited)
Sounds good.
3.98a4 is now at Rarewares. 
This is a "lo-fi" version of our main content. To view the full version with more information, formatting and images, please click here.