IPB

Welcome Guest ( Log In | Register )

 
Reply to this topicStart new topic
frequency domain distortion, some recordings have high levels of all frequencies
AndyH-ha
post Dec 7 2012, 01:06
Post #1





Group: Members
Posts: 2192
Joined: 31-August 05
Member No.: 24222



Has anyone run into this kind of distortion problem? Most especially, is there a solution?

The levels are good, no clipping. Spectral view shows that fairly high, and pretty constant, levels of unnatural sound is present from somewhere around 2kHz to 4kHz, all the way to the Nyquist limit. In reality, the distortion might start somewhere near 0Hz but I can't really tell from what I see.

The recordings are spoken audio coming from cassette. This distortion is only present when speech is present but there is more background noise than usual during the pauses. Most of the real speech content is 4kHz and less, but there is a little up to about 10kHz. When this (intermittent) problem occurs, it is there from the first word on the cassette to the last (in the computer recording). It is not on the cassette. The first time it happened I re-recorded without difficulty. There are now a few more recordings I must do over. I have not yet looked at those cassettes but the problem effected only one side of each. I much doubt that the cassettes, or the recordings on them, are at fault.

When I record, I monitor at the tape deck headphone out. Everything sounded proper there on all of the cassettes. Often, but perhaps not always, I listen to the headphones from time to time as the cassette is playing, just to be sure things are still good, but I suppose that is irrelevant since when the problem occurred it occupied the entire recording session. It never started sometimes during the recording.

What is new is the computer. Actually it is a used corporate machine running WinXP professional that I acquired recently. This newer used computer has a Intel Core 2 Duo running at 3GHz, 2GB of RAM, and two SATA hard drives, running up-to-date WinXP Pro. I installed a Echo Mia soundcard, which I had been using for many years in a much older, slower Win98 computer, with a current (probably several years old, being for WinXP) Echo driver.

The machine isn't allowed to do anything automatic that I can find. I disabled the anti-virus, firewall, task scheduling, etc. Besides, all my experience with excess background activity says missing samples will be the result of too few resources, not frequency domain distortion.

The soundcard is using Echo's Purewave driver, which Echo says addresses the soundcard hardware directly, bypassing all Windows manipulations. I could not use that driver on Win98, but everything worked quite well there. For this latest set of cassettes, where the problem surfaced, I switched from recording at 16 bit to 32 bit (Cool Edit 2000). Prior to the switch to 32 bit, I had recorded more than 20 cassettes on the same setup at 16 bit with no problems.

The weird behavior might be related to the higher bit depth, the problem first appeared on the first cassette, recorded at 32 bit, but since it does not effect most recording sessions, that can't be the complete answer -- if it is relevant at all.
Go to the top of the page
+Quote Post
xTobix
post Dec 7 2012, 02:39
Post #2





Group: Members
Posts: 12
Joined: 1-August 12
Member No.: 101938



Hi AndyH-ha,

it sounds like a problem on the computer/sound card side. I would switch back from 32 to 16 bits and record a part of one of the distorted cassettes again. (16 bit is more than enough for this I would imagine).
If you can reproduce it, you might want to used this http://audacity.sourceforge.net/ (a free audio editor and recorder).

You can use audacity to look at the spectrum.

Good luck!

EDIT1: does your sound card actually support 24 or 32 bits? Judging from the age 16 it probably is. Hence, this could well be the source for the experienced distortions!

This post has been edited by xTobix: Dec 7 2012, 02:46
Go to the top of the page
+Quote Post
AndyH-ha
post Dec 7 2012, 06:53
Post #3





Group: Members
Posts: 2192
Joined: 31-August 05
Member No.: 24222



The soundcard is 24 bit. Like every other 24 bit soundcard, its dynamic range is less than 24 bit, but it works at 24 bit. Thermal properties of basic electronic components prevent converters getting very near 24 bit without cryogenic cooling.

This can not be reproduced at will. So far, no cassette has made a distorted recording twice. It might go away completely if I only recorded at 16 bit but I've been using the soundcard at both 16 and 24 bit (32 bit floating point is the file format) at will, for years with no problem. If the problem resides in the soundcard, and not in the computer or WinXP, it is a problem that just developed.

Yes 16 bit is plenty for the material. It is probably plenty for the processing too but the relative amount of quantization error is markedly smaller for each step when in floating point format, so I choose to use it.

Recording at 16 bit, and converting to 32 bit before other processing, may take less extra time than re-recording the occasional piece, if the higher recording bit depth actually has anything to do with the problem. In terms of actual total time, which is more efficient would depend on how frequently it happens. I already understand all these obvious things, I am just hoping for some matching experience or insight to the actual malady.
Go to the top of the page
+Quote Post
xTobix
post Dec 8 2012, 06:51
Post #4





Group: Members
Posts: 12
Joined: 1-August 12
Member No.: 101938



QUOTE (AndyH-ha @ Dec 7 2012, 06:53) *
The soundcard is 24 bit. Like every other 24 bit soundcard, its dynamic range is less than 24 bit, but it works at 24 bit. Thermal properties of basic electronic components prevent converters getting very near 24 bit without cryogenic cooling.

Agreed - I cool my detectors at work to -162 deg C ;-) Maybe I should do that with my DAC too :-D

QUOTE (AndyH-ha @ Dec 7 2012, 06:53) *
This cannot be reproduced at will. So far, no cassette has made a distorted recording twice. It might go away completely if I only recorded at 16 bit but I've been using the soundcard at both 16 and 24 bit (32 bit floating point is the file format) at will, for years with no problem. If the problem resides in the soundcard, and not in the computer or WinXP, it is a problem that just developed.

If hardware and driver software support 24 bits and you have done it before then I doubt this is the source of the noise.

QUOTE (AndyH-ha @ Dec 7 2012, 06:53) *
Yes 16 bit is plenty for the material. It is probably plenty for the processing too but the relative amount of quantization error is markedly smaller for each step when in floating point format, so I choose to use it.

I don't know how much post processing you are going to apply. I would say with a regular cassette player and tapes of some age you are somewhere between 10-14 bit before it disappears in the noise floor. However, this is entirely up too you.
Since it comes and goes this has to be some noise source close or inside your PC. My first attempts to tackle this would be in order:

1) Run all you fans including the one in the power supply itself of an external power supply/battery. ( you can use a second PC power supply just for the fans!

2) turn off Bluetooth, WiFi, disconnect LAN, and disable all other soundcards (onboard in bios)

3) Remove all drives and cards not essential for recording from the power supply and the data bus (i.e. 2nd/3rd HDD, CD-ROM, DVD, Floppy drives, and all other PCI cards). Also all external devices attached on USB, RS232, LPT, firewire, ... (printer, drives, ect.)
4) Take some aluminum kitchen foil and try to shield your entire soundcard (make a sort of sleeve that slides over the pci card). You have to be very careful that you do not shortcut anything (you could use some plastic sleeve first) and the aluminum must be in good contact with ground!
5) Terminate/Short cut all unused internal and external inputs and outputs with a wire or some plug.

To me, it seem like a fan is turning on and off or a drive or something else. Fans are nasty noise sources by their nature (this is true for DC and PWM driven fans!).

Let me know if this solved/change things.
Go to the top of the page
+Quote Post
AndyH-ha
post Dec 8 2012, 20:56
Post #5





Group: Members
Posts: 2192
Joined: 31-August 05
Member No.: 24222



This is not added noise, it is distortion, occurring somewhere in the recording chain.
I only mentioned the 24 bit thing because I have no prior experience with Windows XP audio; I don't know that there isn't something peculiar required to get WinXP to behave properly.

The last entry in this thread
http://www.hydrogenaudio.org/forums/index....showtopic=54879
contains a sample

As always, this unique way that HA has of making it difficult to attach something to a post, makes things awkward and inconvenient, but it is there. The audio sample has been converted to 16 bit so that my slow connection could get it there is a more reasonable time. You'll just have to take my word for it that the conversion did not change anything significant and that none of the awful stuff is from the cassette.

This post has been edited by AndyH-ha: Dec 8 2012, 21:03
Go to the top of the page
+Quote Post
Dynamic
post Dec 9 2012, 19:25
Post #6





Group: Members
Posts: 793
Joined: 17-September 06
Member No.: 35307



The spectrum indicates fairly regular peak and trough features, like a comb filter of sorts, that are partly evident in AndyH-ha's upload forum spectrogram

Not sure what to make of it
Go to the top of the page
+Quote Post
AndyH-ha
post Dec 9 2012, 20:00
Post #7





Group: Members
Posts: 2192
Joined: 31-August 05
Member No.: 24222



As I said in the beginning, this is speech only, although I have no reason to believe that the same thing could not have happened had I been recording music. In this material, breaks between words, and pauses between phrases, sentences, etc., are the norm. The strong high frequencies are only, and always, there during words.

Bands of noise across the spectrum are visible in the pauses, much fainter, but undoubtedly related to the problem since they do not appear in the normal recordings, which show only more or less evenly distributed tape hiss..
Go to the top of the page
+Quote Post
lvqcl
post Dec 9 2012, 20:36
Post #8





Group: Developer
Posts: 3212
Joined: 2-December 07
Member No.: 49183



There's something wrong with data processing IMHO. Maybe a bug in the drivers?
Go to the top of the page
+Quote Post
AndyH-ha
post Dec 10 2012, 04:39
Post #9





Group: Members
Posts: 2192
Joined: 31-August 05
Member No.: 24222



Drivers for XP haven't been updated for a few years. If there was a bug, it probably would have been addressed while XP was in its prime. Still, when I don't know anything, everything is possible. Apparently no one around these parts has seen anything of this sort.
Go to the top of the page
+Quote Post
Dynamic
post Dec 10 2012, 17:04
Post #10





Group: Members
Posts: 793
Joined: 17-September 06
Member No.: 35307



QUOTE (AndyH-ha @ Dec 9 2012, 19:00) *
As I said in the beginning, this is speech only, although I have no reason to believe that the same thing could not have happened had I been recording music. In this material, breaks between words, and pauses between phrases, sentences, etc., are the norm. The strong high frequencies are only, and always, there during words.


I was meaning not in the time domain of the spectroGRAM, but in the frequency domain alone the spectrUM analysis over a given interval during the speech sounds, there's a lumpy spectrum (see my plot in the Upload thread) during the distorted parts, related in some way to the signal. (This also shows up less clearly in the spectrogram as horizontal lines of consistently lower energy in the frequency domain). Maybe there's a time-delayed distorted version mixed with the straight signal, which would cause the spectral peaks and troughs at regular frequency intervals (troughs at odd multiples of about 700Hz as I recall) as the two signals are added. It doesn't appear to be in any way anything like big-endian or little-endian swapping.

Still, I don't think it's getting me anywhere as I can't make a good guess as to what it is, especially without an undistorted version to compare to.
Go to the top of the page
+Quote Post
2Bdecided
post Dec 10 2012, 18:03
Post #11


ReplayGain developer


Group: Developer
Posts: 4945
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



Could it be some really atrocious and unwanted attempt at sample rate conversion somewhere? e.g. a driver dropping or duplicating samples? or performing nearest neighbour interpolation or something?

You can always find a bug on a specific combination of hardware, software and driver versions that no one put together before.

Cheers,
David.
Go to the top of the page
+Quote Post
Dynamic
post Dec 10 2012, 20:26
Post #12





Group: Members
Posts: 793
Joined: 17-September 06
Member No.: 35307



QUOTE (2Bdecided @ Dec 10 2012, 17:03) *
Could it be some really atrocious and unwanted attempt at sample rate conversion somewhere? e.g. a driver dropping or duplicating samples? or performing nearest neighbour interpolation or something?

You can always find a bug on a specific combination of hardware, software and driver versions that no one put together before.

Cheers,
David.


Actually, a close up of the waveform shows dramatic fairly regular jumps, and apparently repeated shapes roughly every 30 samples.

See this post in the Upload thread for an example.

Having two copies of the same shape might even be some dodgy attempt to handle stereo channels and downmix to mono.

It looks as though many parts of the waveform are missing and replaced by a repeat of a previous part, and that this is done roughly every 30 samples.

This implies peaks at about 1470 Hz and multiples and troughs at about 735 Hz and odd multiples (i.e. between 1470) which matches the apparent comb filtering I saw.

Very peculiar.
Go to the top of the page
+Quote Post
AndyH-ha
post Dec 10 2012, 22:13
Post #13





Group: Members
Posts: 2192
Joined: 31-August 05
Member No.: 24222



This pattern is apparent, and possibly some clue about the cause, but it isn't normal harmonic distortion, which tends to be at a significantly lower level for each additional step. Isolating any particular frequency band, either by the peaks or troughs, as revealed in a frequency analysis, does not improve the intelligibility any; all of the recording seems to be quite distorted. There is no more-or-less undisturbed source with distorted duplication at higher frequencies.

The material is mono but there are two tracks of it on the cassette. I record both and select whichever seems best. In general, either would do, they do not vary greatly. In these cases, both seems as bad.

I detect no trace of the problem in most recordings, and no trace in the re-recordings of these particular cassettes. Whatever it is that happens, it starts with the beginning of the recording session and persists to the end. It does not, so far, ever happen with the next recording session, even if that is simply the short time necessary to turn the cassette over, start it playing, and press Record again.

The Echo drivers are intended to bypass all Windows audio manipulations. There should not be any resampling happening.
Go to the top of the page
+Quote Post

Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 19th April 2014 - 17:01