Asychronous Blocks!

2002-09-14 19:54:24

My understanding is that low frequencies have an intrinsically poor time resolution and high frequencies intrinsically good time resolution. Why then do we impose the same window on all frequencies during Fourier analysis??

Wouldn't the most sensible thing be as follows:
1. Analyse the waveform with SEVERAL Fourier analyses, with each having a different window size.
2. Get information about the high frequencies from the analyses with short windows (and discard their take on the low frequencies); get information about the low frequencies from the analyses with long windows (and discard their take on the high frequencies!)
3. Match the blocks in time

This way the high frequencies always have short blocks (eliminating ringing) while the low frequencies have longer blocks to preserve tonality!

Or actually you can choose to take the results for whatever frequency from what ever window length Fourier analysis you want--if for some reason you want higher frequency resolution for the high frequencies as well. Anyway this should increase flexibility greatly!

I'm sure there must be something wrong with this, otherwise somebody would have done this by now

Asychronous Blocks!

Reply #1 – 2002-09-14 20:46:30

That is what the MP3 specs allows.. The lower freqs are calculated from long block while the upper part of the spectrum comes from short-blocks.
However, I do not understand why the AAC specs dumped this implementation.. It switches to 8 short-blocks completely.. Even the MP4 SSR profile filterbank does not make use of this concept... Perhaps, it is because of the TNS tool ? Also I noticed that for some clips such as "fatboy", it may be necessary to completely switch to short-block for the entire spectrum..

wkw

Asychronous Blocks!

Reply #2 – 2002-09-14 20:50:25

eh? Based on this you would expect mp3 to handle pre-ringing better...

BTW, this seems to be similar to the wavelet concept. (I just found out  ) What is there that wavelets can do that asynchronous blocks can't?

Asychronous Blocks!

Reply #3 – 2002-09-14 21:10:39

Is pre-ringing means pre-echo?
If so, pre-echoes normally occurs at the upper part of the spectrum. So, I guess, for most pre-echoes, it may not be necessary for the lower part of the spectrum to switch to short-block.. The Gain Control Tool for MPEG4 AAC SSR ignores  the lower freq band as well.

I know very little about wavelets.. My experience in video / image compressions are limited to DCT compressions..

wkw