Audacity and Media Player on Windows 7 notebooks

[Gale Andrews]

Great tip to double-click the sliders for fine control.

... I would agree this isn't really expected. What else could be done in the space to make it more discoverable isn't so obvious. It would be a bit tiresome to say "double-click for fine adjustment" in the tooltip, I think. Right-click has to be used to select the slider at the moment. ...

On Linux, the "current dB level" tool-tip only pops up when the user clicks on the volume slider.
Other tool-tips pop up on hover (for example, the record button shows "Record (Shift for Append Record)" when the mouse is over the button but not clicked. Is it possible to have both types of tool-tip on the same control? If this is possible, how about keeping the dB display tool-tip as an "on-click tool tip" and "double click > fine adjustment" as an "on-hover tool tip"?

The main problem is that a lot of users (rightly IMO) beef about not being able to to discover the value the slider is at without clicking carefully on the handle (if you click elsewhere you move the slider...). I don't think that's acceptable for something like gain/pan sliders where ticks are at more than one unit. Then on some systems, the two different tooltips obscure each other. See bug 74.

+1 for configurable clipping level. I have experienced a sound card that would always clip below 0dB no matter what settings were used. Some users may also find it helpful/preferable to set their clipping indicator to less than 0dB even if their hardware is not faulty simply to give them a visual indication of when the level is close to clipping. (they can always reset the indicator to 0 dB to check if it actually has clipped.

I'll add your vote to the Wiki Feature Requests so it doesn't get lost. However I'm not sure having a variable level for the clipping hold light is a great idea unless it's obvious what it's currently set at. Maybe the peak (blue) indicator should change colour when the chosen level is reached?

-1 for configurable clipped sample runs. Whether it is only 4 samples that are clipped, or more than 4 samples that are clipped, it should be avoided. Allowing a greater degree of clipping without showing any warning will only lead to more user problems with distortion.

The main rationale given by those who voted for this is a) they want the clipping triggers the same for all three methods of displaying it; b) they are more comfortable with the meter indicating clipping the first time one clipped sample is encountered. So the idea would be (if the meter remains being triggered at four consecutive clipped samples) that a triger level of 3, 2 and 1 consecutive samples would be offered (but not more than that). Considering the statements I've seen that Find Clipping was set to a threshold of three in order to reflect the meter trigger level, it could be the meter trigger level isn't even what is intended.

Volume 0 (34 Mb): http://www.sendspace.com/file/0xjgho

The clipping on this sample is worse than indicated by "show clipping". In some places "show clipping" indicates a red line for one sample, but the next few samples are clearly clipped, but at slightly less than 0 dB - as can be seen here at 39.1170.
The first sample that shows clipping is at 0 dB which is the limit of 16/24 bit audio, but the next 14 samples are clearly clipped, but are at just under 0 dB, which would tend to indicate that the clipping is due to the sound card being overloaded.

Thanks, Steve for looking more closely. There wasn't much "flat-topping" (like a sandcastle, which is the visual indicator of clipping) where I was looking in the loud section, but this must be the basic explanation why the clip sounds so bad with such limited consecutive clipping. It's also fair to say looking at the "almost unclipped" version that a number of the peak areas would look suspiciously "like" clipping if they happened to be closer to +/-1.

A possible problem convincing developers that the meter trigger should be one sample may be that arguably this audio would still sound bad even if it had been recorded a bit lower, so that only one sample at a time was clipped.

On Linux it is possible (though bad practice) to record over 0 dB even from a 16 bit sound card, provided that it is recorded at 32 bit. I presume this is due to the 16 bit audio being scaled beyond 0 dB if the recording levels are all set to maximum. (Alsamixer "red-lines" at 72%)

So in this case you still lose dynamic range over the theoretical case where the input was 32-bit?



Gale

[steve]

So in this case you still lose dynamic range over the theoretical case where the input was 32-bit?

I'm not sure that I follow you?

[Gale Andrews]

So in this case you still lose dynamic range over the theoretical case where the input was 32-bit?

I'm not sure that I follow you?

I wasn't sure that I followed you about the scaling. I thought perhaps you were suggesting that the samples would be less above +1 where you were recording a 16-bit device at 32-bit float, than if you really were recording a 32-bit device at 32-bit float.



Gale

[R_G_B]

OK, what Gale and Steve are saying may finally be sinking in.

If I have it straight, it is important for other users, because:
1. These days most computers sold are notebooks (or netbooks).
2. It is lately nigh impossible to find a notebook with a Line In jack.
3. The Griffin iMic 2 is probably typical of reasonably priced usb sound cards (it even gets a mention in Audacity wiki).

So, from Steve's advice about ideal recording levels in digital recordings (Audacity) as distinct from the old days of analog tape; we should aim to have most loud sections not just below 0 dB, but rather around -6dB (equivalent to height of 0.5 on the vertical scale). Probably iMic is already clipping when fed a Line signal in the Mic position (as Gale was also saying). I tested it in 'Take 3'. So the advice would be that the top panel (iMic in Mic position with Audacity record slider at 0) was recorded too loud, and the middle panel (iMic in Line position with Audacity record slider at 100) is the way to go. The Take 3 project zip is at: http://www.sendspace.com/file/lk9o7h

imic.jpg (72.83 KiB) Viewed 2220 times

Should we be worried by the extra noise from the iMic (with no input signal) when the recorder slider is pushed from 0 to 100 as shown in the bottom panel?


The computer Mic in source also shows more noise at 100, though without the wave pattern:

Mic.jpg (12.08 KiB) Viewed 2220 times

[waxcylinder]

...The Griffin iMic 2 is probably typical of reasonably priced usb sound cards (it even gets a mention in Audacity wiki).

Some of our posters have had problems with the iMic in the past. One of the forum elves, Kozikowski, had one - and junked it to his garage. Stevethefiddle uses the Behringer UCA202 and I use the Edirol UA-1EX.

See this sticky thread in the forum: http://forum.audacityteam.org/viewtopic.php?f=27&t=9477

WC

[steve]

Should we be worried by the extra noise from the iMic (with no input signal) when the recorder slider is pushed from 0 to 100 as shown in the bottom panel?

Unfortunately yes, we should be worried.
I think the problem is all about where the manufacturers cut corners.

As WC has just said, I use a Behringer UCA-202. The obvious place where Behringer have cut a huge corner is that there is no recording level adjustment (it's also built into a cheap plastic box and only provides 16 bit recording). The recording level is fixed at "line-level" (fortunately it's a fairly generous "line level" with a specified max peak level of +2 dBV and so avoids any distortion from consumer audio equipment "Line" connections).

On the other hand, the Edirol has cut less corners and provides input level adjustment and very good sound quality, but the price is quite a bit higher.

The iMic is attempting to provide the features of the Edirol, but at the price of the Behringer, and we are seeing (hearing) the cost of that. It seems that the input sensitivity when set to "Line" is just too low (hence the noise), but too high when set to mic (hence the distortion).

As I see it, the options are a bit limited:
1) Record at Line level and put up with the noise (may be able to improve it using the Noise Removal effect)
2) Record with it set to mic and put up with the (slight) clipping distortion.
3) Modify, or add a resistor divider circuit to reduce the sensitivity when set to "Mic".
4) Upgrade to a better device.

If you can handle a soldering iron and want to try building a simple resistor divider circuit, I can give you details of how to do that. Otherwise, I think you will need to take your pick of one of the other three options.

Anyone else got any other ideas?

I wasn't sure that I followed you about the scaling. I thought perhaps you were suggesting that the samples would be less above +1 where you were recording a 16-bit device at 32-bit float, than if you really were recording a 32-bit device at 32-bit float.

I'm not quite sure why/how this happens, but if I crank up all the levels in the computer, I can record a peak, undistorted level up to +6 dB. (tested using the internal 16 bit sound card). On playback it sounds badly distorted, but amplifying by -6 dB will bring it back to within the 0 dB range and it will then play without distortion. If any peaks go beyond +6 dB, they are clipped at +6 dB and cannot be recovered by negative amplification.

[R_G_B]

Thanks WC and Steve, I have revised points 1 and 8 in the initial post to incorporate your sage advice (struggling to incorporate the key points within a 2-page doc).

The background noise from the iMic at input slider 100 depends on the connected wiring. It looks bad with a blank cable plugged to the in jack. The four sections recorded are Line 0, Line 100, Mic 100, Mic 0.

iMic-noise.jpg (36.16 KiB) Viewed 2205 times

The noise is worse if you put a hand near the iMic, but much less if the blank cable is detached from the iMic. Listening to the last few seconds of the previously posted recordings it is hard to distinguish above vinyl noise - user beware I guess.

Steve, I can solder. But I am still thinking about that apparent hardware clipping. Zoomed in I could still see it at the loudest point in the Take 3 project (iMic Line with slider 100). I guess I must record again with the input slider set lower to be sure that this peak was not actually clipped in the Sheffield direct to master recording. Otherwise we may blame iMic wrongly.

Jumping ahead, would the resistor divider have to go on the input side of the iMic if the amplifier Line out is indeed being hardware clipped in the iMic?

P.S. I have left the warning about 'failure' of the Audacity meter clipping alert, so that dumb users (like me) know it is not 'safe' to simply push up the input slider provided there is no clipping alert on the meter. I take Gale's point that checking for other triggers (such as 100 clips in any second) might use too much processor. How about setting it at any four consecutive points above +0.95 dB? It looks like that might alert users about excessive record volume even if their soundcard is intervening to keep levels just at or below 0 dB when the slider is pushed too high. From Steve's lesson, nobody should want to record in digital at that level.

[Gale Andrews]

Considering the statements I've seen that Find Clipping was set to a threshold of three in order to reflect the meter trigger level, it could be the meter trigger level isn't even what is intended.

The meter trigger on close inspection *does* seem to be working correctly at three consecutive clipped samples, so matching the Find Clipping default. Sorry for any false alarms. In Audacity you can drag a region starting from any point in the waveform between or on the sample dots, and although this exact region is shown on the timeline and reflected in Selection Toolbar, the region in the waveform snaps between the sample dots and doesn't match the timeline region. So as I was doing, you can select across three samples in the waveform while in fact you are only selecting two samples. This isn't of course obvious unless you are zoomed in as far as you can go.

In terms of the level at which clipping is detected it seems to be at about +/-0.99997 dB for Meters, Show Clipping and Find Clipping, so no discrepancy there (I haven't attempted to look at the actual code). There is then just the question if the Meters should have an optional lower trigger level and if the number of consecutive clipped samples to trigger should be only one or if an option for that should be provided.

Some good news with your Take 3 done with line-in is that the 1.3 Noise Removal seems to do a good job of removing the combined vinyl and iMic noise even at default settings.

How about setting it at any four consecutive points above +0.95 dB? It looks like that might alert users about excessive record volume even if their soundcard is intervening to keep levels just at or below 0 dB when the slider is pushed too high.

I'll try and get around to raising a review bug about trigger levels and numbers and then the developers who know about signal processing can consider it. Looking at the first badly clipped audio you sent, it looks as if some of the clipping might start as low as -0.8 which I think is too low for a default clipping indication. However there is always the danger that a beginner could record something irreplaceable in your case, see no clipping on the recording meter, not bother to listen to it, export it and not realise there is a problem until too late.

The problem you have when recording line-level to mic level wouldn't really be called "soft clipping", but another issue that has sometimes been talked about is if the meters should try and detect soft clipping which could sometimes be a fault and sometimes be deliberate, according to the device.


Gale

[R_G_B]

In fairness to iMic, there was something strange about the first cable I tested for background noise. It may be picking up RF (worse when I touch the cable). Other tested cables (with or without ferrite beads) do not do this. They give much lower noise levels, similar to the iMic without an attached cable. Again the four sections recorded are Line 0, Line 100, Mic 100, Mic 0. Odd that R is quieter than L, but both are <.005 at 100 on the input slider:

imic-ferrite.jpg (36.47 KiB) Viewed 2182 times

That helps to explain why the noise is not noticeable in recordings (which did not use the 'problem cable'), even with the Line / 100 settings, and it can be filtered further as Gale says.

After all the advice from Gale, stevethefiddle and waxcylinder, the recordings from LPs are much better than you might expect using a basic notebook with a sub-$40 usb sound card!
The information from stevethefiddle about lower recording levels for digital versus tape was a great lesson to me.
With any luck, the tweaks along the way to incorporate their advice into the initial 'short form tutorial' might help other occasional users to get to the same great results quickly with Audacity on notebook computers.

Much obliged. RGB.

[kozikowski]

We are seeing multiple real-world problems.

-- Rejection of computer USB battery noise is expensive and generally left out of cheap equipment. This is the "I can hear my hard drive spinning up in my show" problem. This is one reason my iMic is in the car park.

-- Rejection of radio interference is expensive and generally left out of cheap equipment. I can tell precisely when one of the editors is going to get a cell call. I can hear it in my speakers (not the actual conversation).

-- Microphone level is reeeeely, reeeely tiny. There are common electrical components whose natural noise (thermal, shot, etc.) is the same as mic level. Quiet components are expensive and generally left out of cheap equipment. One noise shortcut you can take is to make the first amplifier in a microphone system relatively cheap, quiet, and fixed. You can't change it -- ever.

This works fine as long as your signal is actually microphone level. This is the amplifier that overloads when you apply a thousand times too loud line-level signal and it's what causes people's Audacity show to clip at 0.5. It can also clip at 0.99 (or any arbitrary value) which can give you some odd clipping data. Even cheap line-in electronics can do this. Everything is designed to run from the same five volt USB or computer battery. You can't make higher voltages, either, because that cost money.

It is required that the analog system overload at a higher value than the digital for the digital overload tools to work and they don't always conform. The only way I know of to make sure which system is falling over first is with instruments that measure the signal on the sound card. Also, oddly enough, expensive.

Koz