Post reply

Warning: this topic has not been posted in for at least 120 days.
Unless you're sure you want to reply, please consider starting a new topic.

Note: this post will not display until it's been approved by a moderator.

Name:
Email:
Subject:
Message icon:
Verification:
Sum of two plus two?:

shortcuts: hit alt+s to submit/post or alt+p to preview


Topic Summary

Posted by: Admin
« on: October 04, 2010, 00:07:01 »

Thanks for you comment,
We originally planned to add rating to duplicates results, but because it can used more generally, we add firstly it standalone, after sometime and test we integrate them more deeply.
Posted by: hsei
« on: October 03, 2010, 12:04:51 »

OK. besides the VBR problem there is another issue:
At the moment the quality analysis is done for all files in the folder list. I would find it more appropriate to do that only for files in the results list (candidates for similarity). That would reduce processing effort drastically in most cases.
If the result list is empty, you may still perform an analysis on all files. That's a nice feature on it's own - but not within the original scope of removing almost identical tracks (keeping those with higher quality).
Posted by: Admin
« on: September 26, 2010, 11:51:46 »

Issue with VBR bitrate fixed, solution more heuristic - we take raw bitate filesize/duration and use it, if it deviate from codec bitrate more than 10%. And we corrected duration, we now take it from samples count and sample rate, not from codec info.

Abou FFT, we planning to add spectrogram with avg./max. frequencies, but we don't decided yet.
Posted by: hsei
« on: September 24, 2010, 19:20:37 »

According to frequency bins:
I have found at least another frequency indication of 22028 which means you have bins of not more than 22 Hz range (1024 FFT). Important is not if there is any contribution in that bin but to what extent. In analog to digital conversion a proper low pass filter prior to sampling at 44.1 kHz should start at about 20 kHz and suppress frequencies at the Nyquist limit by at least 30 - 50 dB. An indication of a cutoff frequency where all contributions above have fallen below a given limit of some ten dB is much more meaningful than the maximum bin with some isolated spurious (and negligible) content.
Compressions to low bitrates usually use filters with lower cutoff frequencies but may show minor artificial high frequency content after decompression due to incomplete restauration of the original signal. By just looking at the maximum bin you would rate these files too high.
Posted by: hsei
« on: September 24, 2010, 17:28:49 »

OK, I see your problem. Similar behaviour appears in the indications of Microsoft Explorer. They are sometimes completely wrong for bitrates (not only for VBR). You cannot trust these data for sensitive operations like deleting the "worse" duplicate.
Posted by: Admin
« on: September 24, 2010, 15:33:35 »

the problem in decoder, we not decode it ourself, we decode it via WMF decoder and it send us such information. We think about correcting such information.

about freq., yes you right, but FFT shows not value on frequency, it show value of the freq. range (frequency bin), we show the right most in range in the last bin, but yes, this value not achievable, all values less than it.
We can rename it from "Max. freq." to "Lim. freq."
Posted by: hsei
« on: September 24, 2010, 11:09:14 »

The new analysis mode is a nice feature, but has to be used with caution:
VBR files are rated totally false, the bitrate is taken from the first frame(s) where it is typically very low. Consequently the duration is calculated much too high (total size by first frame size).
Reading of further frames seems to take into account the bitrate in frame header, otherwise complete rubbish would appear.
I would advise to ether correct ratings for VBR to average bitrate or drop rating (at the moment)  for VBR at all.
In frequency analysis I found another problem: I got a rating of 100% for a file sampled at 44100 Hz with a max. frequency of 22050. Files should never contain frequencies up to the Nyquist frequency (half the sample rate), because that leads to annoying alias frequencies. That should lead to drastic reduction in quality rating. For the file in question I found no perceptible degradation, so the max. frequency value may be an analysis flaw.
Again: Be reluctant with analysis ratings, but with some caution it's a valuable feature. Calculating quality measures is a demanding task.