Quality problems when converting from FLAC to MP3 LAME

[breedingbull88]

Hi,

I have picked a flaw in the application.

When converting from FLAC (FLAC saved from Vinyl recorded at 32bit floater @ 48Khz), to MP3 (lame) 320kbps, you have the option to pick the option (under the drop-down called ENCODING).

* Slow (High Quality)
*Normal
*Fast (Low Quality)

It seems when I convert using Slow & Normal, the quality degrades (speks at higher frequency fades away) when you run the Spectal frequency Chart (using Adobe Audition 3). However when I encode using "Fast", and analyse the frequency, the quality is retained far better than Slow & Normal? The speks are more fuller and richer at higher frequencies (around 20,000Hz).

Surely there must be a categorisation error where Slow was meant to be "Fast" and Fast is meant to be categorised as "Slow"?

I am running LAME 3.99 (latest version).

[Spoon]

Re: Quality problems when converting from FLAC to MP3 LAME

Spectral Plots show nothing about quality, otherwise LAME would preserve right up to 22KHz.

[zipzip]

Re: Quality problems when converting from FLAC to MP3 LAME

Hi,

I have picked a flaw in the application.

When converting from FLAC (FLAC saved from Vinyl recorded at 32bit floater @ 48Khz), to MP3 (lame) 320kbps, you have the option to pick the option (under the drop-down called ENCODING).

* Slow (High Quality)
*Normal
*Fast (Low Quality)

It seems when I convert using Slow & Normal, the quality degrades (speks at higher frequency fades away) when you run the Spectal frequency Chart (using Adobe Audition 3). However when I encode using "Fast", and analyse the frequency, the quality is retained far better than Slow & Normal? The speks are more fuller and richer at higher frequencies (around 20,000Hz).

Surely there must be a categorisation error where Slow was meant to be "Fast" and Fast is meant to be categorised as "Slow"?

I am running LAME 3.99 (latest version).

While Spoon is correct that rolloff of higher frequencies doesn't necessarily mean anything from an audible quality standpoint, I have been discussing this very issue on Hydrogen Audio Forums: that the higher "quality" settings in LAME actually degrading the sound. They tell me it is an issue that has been known about for some time. The well-known "killer sample" called "Show Me Your Spine" actually shows noticeably MORE artifacts when using the "high quality" setting than when using the "low quality." After futzing around with it for literally hours, making different samples, different settings, I finally have reached what I feel is the likely cause, though I can't prove it:

My theory is that the quality settings on LAME were only tested by LAME's developers on Variable Bitrate files. I speculate they did very little testing at the Constant Bitrate setting. Someone else had brought up this possibility on Hydrogen Audio Forums, but I later was able to verify that the "Show Me Your Spine" sample showed fewer artifacts when using the high quality setting -- but only when I was encoding with variable bitrate. On a flat 320 CBR file, just the opposite was true, for reasons no one seems to know. In a nutshell, the quality settings may not have been designed for constant bitrate, only variable birate files. LAME defaults to Q3 (should match dBpoweramp's "Normal" if no quality value is listed in the commandline.

In my opinion, you should probably keep it at "Normal" for 320kb files, which I believe is LAME quality setting of Q3. At "Fast (Low Quality)," which I think is equivalent to Q7, what happens according the LAME detailed commandline switches documentation is that it does away with "noise shaping," but retains LAME's psy-model for pre-echo and m/s stereo.

http://lame.cvs.sourceforge.net/viewvc/lame/lame/doc/html/detailed.html*q

At any rate, it is NOT a problem with dBpoweramp, rather with LAME.