The Relation between Time res. and frequency

Topic: The Relation between Time res. and frequency (Read 3579 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.

The Relation between Time res. and frequency

2003-12-23 14:22:45

Hello

Sorry to be so ignoramus (this name tells me something... Radi... ) but i want to kown is there is a bond between the time resolution and the frequency of a sound.
IHMO increasing the last improves the first in that way you can reduce the pre-echo artefact.
My question is very simple: is it better to convert to 48 kHz before encoding ? (even with codecs what use 192 samples for short blocks).

Thank you for your precious help.
Have a merry christmas !

Sincerely,
Nick

Edit: I'm sorry but as you can imagine, the title is not "the" but "Relation between Time res. and frequency".
Sorry

The Relation between Time res. and frequency

Reply #1 – 2003-12-23 15:02:08

Quote

My question is very simple: is it better to convert to 48 kHz before encoding ? (even with codecs what use 192 samples for short blocks).

There's been some discussion here about this. IIRC the results were this:

mp3 (lame)
- lame (especially alt-presets) is tuned for 44.1kHz and performs audibly worse on some samples at 48kHz (but not better anywhere IIRC).
- pre-echo could be reduced somewhat using 48kHz theoretically, but OTH there are more frames/blocks per second, so more bits are needed to encode the same - that's why for low bitrates resampling to e.g. 36kHz is recommended.

musepack
There are a few problem samples that improve using 48kHz.

The Relation between Time res. and frequency

Reply #2 – 2003-12-23 15:29:21

There is a relation between the time resolution t and the maximum frequency Fmax of a sound :

Fmax=1/(2t)

But the pre echo artifact in encoding don't directly depend on this.

Quote

- lame (especially alt-presets) is tuned for 44.1kHz and performs audibly worse on some samples at 48kHz (but not better anywhere IIRC).

Can you find a link ? I know that resampling to 48 kHz is not recommended, but I don't remember about tests showing problems.

The Relation between Time res. and frequency

Reply #3 – 2003-12-23 17:59:57

Quote

Quote
- lame (especially alt-presets) is tuned for 44.1kHz and performs audibly worse on some samples at 48kHz (but not better anywhere IIRC).

Can you find a link ? I know that resampling to 48 kHz is not recommended, but I don't remember about tests showing problems.

Here I found:

Quote

<snip> (DVD 48kHz PCM track -> mp3). <snip> I found two samples in the track where I could successfully ABX 48kHz lame 3.90.2 aps vs. Original but not 44.1kHz lame 3.90.2 aps (downsampled with SSRC before encoding, 24bit output file). Unfortunately I don't have the original anymore, so I can't provide samples. The difference was some slight ringing/chirping added. In spectral view some "dropouts" (black "holes") at ~ 15kHz were visible at these positions (I did this after ABXing ).

I don't know anymore which songs/positions I ABXed. (It was from this DVD, a classical/world music live concert.)

Now I've done a new test using Desmond_1 sample (from 24/96 vs CD quality sample suite):
Original vs.
Original, converted to 44.1kHz/24bit with fb2k resampler (slow) + diskwriter, encoded with lame 3.90.3 --alt-preset standard and
Original, converted to 48kHz/24bit with fb2k resampler (slow) + diskwriter, encoded with lame 3.90.3 --alt-preset standard

Focussing on 1-4s I was able to ABX successfully [Original vs. 48kHz mp3] and [44.1 mp3 vs. 48kHz mp3] but not [Original vs. 44.1kHz mp3]. (The difference I heard was fluttering/bumping of the cymbals/hihats).

To find other similar samples, using music with lots of high pitched, noise-like content should be promising (e.g. live recordings, metal with distorted guitars + cymbals).

Notice