Hi all,
Just wondering, does the sample rate of the source file affect Lame's encoded quality?
For example, if you use a wav file which is sampled at 44k as an input, and you encode at, I dunno, let's say -v5 --vbr-new, will the quality be the same, as if the input wav had first been SRC'd to 32k, and THEN encoded at the aforementioned rate?
Now, I realise that the 44k wav has a frequency response greater than the 32k version, I get that much.
Actually, no, let's start this all over again.
Let's say I start with a 32k wav file. That way, the FR is finite.
Now, let's suppose I SRC the 32k wav UP to 44k.
Both files have the same FR.
But, will the quality of 2 MP3's encoded from these 2 wav's be the same?
I hope I've expressed myself clearly enough.

Higher sample rate means less bits available to actually encode the sound. You'll get better freq. response, but typically worse quality, especially if you're not changing the default lowpass setting.

?

Right!
I had a feeling there was something like that at play, but wasn't sure.
Ta!

We're not talking about 64kbps CBR here. How can there be less bits available?

At a higher samplig rate, the samples are closer together in time, which means that at xx kbps (kilobits per SECOND), those encoded BITS have to cover more samples in less time, therefore lower quality.

But the bitrate isn't fixed. --vbr-new, remember?

Oh, ok... good point.
Well, back to my original question then...

Bitrate isn't fixed, but it's targeted nonetheless. Presets are optimized to aim for a specific quality. If you alter the sample rate and increase it, it will have less room to store what it's looking to store.. Logically.

I guess Gabriel would be best advised to answer this...

OK, that seems fair.
Thanks for your input, Shade.

No, you're making wrong assumptions.

ttf32.wav.mp3 152kbps.
ttf32-44.wav.mp3 157kbps.

OK, now I'm confused.

Hang on, what you're saying there is that the 44k version (which was upsampled from 32k) used a slightly higher bitrate?
Have I got that right?

Right. Overhead, perhaps?

The psychoacoustic model (or the parameters for it) are probably different for different sample rates. I think that this could in principle cause differences quality-wise too, if the fine-tuning of the model was done only for a certain samplerate(s). Probably not the case with Lame, though.

Things are not that easy:

*Reducing sample rate is reducing temporal resolution, as frame sizes are constant regarding number of samples. Reducing sample rate could then increase temporal smearing.

*Reducing sample rate is increasing frequency resolution of the encoded frame/psymodel. Reducing sample rate could then increase possible compression (considering a similar usefull bandwitdh).

*Mp3 is quite inneficient in the upper part of the spectrum, due the a technical limitation. Lame is first selecting lowpass (based on compression ratio), then selecting samplerate in order to try avoiding this limitation.

Advice:
Change the lowpass if you want, but let Lame choose itself the appropriate sample rate based on the provided lowpass.

That reminds me of this topic. http://www.hydrogenaudio.org/forums/index....showtopic=26726

Gabriel. Is it worth upsampling to 48kHz before encoding to avoid the limitations of the MP3 format with high frequency material and improve temporal resolution? I would imagine not, but I don't know.

Regarding avoiding sfb21 limitation (design limitation of mp3), the answer is clearly no, as the sfb21 is approximately starting at the same frequency in 44.1 and 48kHz modes.
Regarding improving temporal resolution: perhaps

They should be close in audible quality. ABX to find out. Average bitrate should be higher for the higher samplerate though. And you expressed your question perfectly clearly to me.

Thanks gameplaya!