Noise Removal in Audacity - Frequency Smoothing

W

winkosmosis

New member
Hey folks, I'm trying to learn how to record for a story filming project I'm working on. I did my first test recordings today and I'm trying to figure out noise reduction. I'm using a Tascam DR-40 and Behringer C-1 with Audacity for cleaning up the audio. I'm recording to 96khz 24bit BWF. As a sound card I'm using a Tascam US-144 set to 96khz/24bit output with some cheap Sony ZX-100 on-ear headphones.

I've been messing around with the noise removal effect and I don't really understand the results I'm getting.

Every guide I find says to use a Frequency Smoothing setting of 100-500 depending on the nature of the noise, and that hiss (which I have) requires a higher frequency smoothing setting. As I understand, when Get Noise Profile on a "silent" piece of the recording, the program figures out the frequencies and gets rid of those from the entire track. And hiss is supposed to be a broad range of frequencies so you need a higher tolerance. But I've experimented and I hear zero hiss even with frequency smoothing set to 0. How can that be? Am I just not hearing the noise? Or is Audacity just good enough at removing hiss that I don't need any smoothing?

Also, I tried different dB settings and I can still hear noise at 10dB and 20dB reduction, so I keep it at 30dB. The 0 setting for Sensitivity seems to work best too. If I change it in either direction I get unremoved noise or distortion. So I think I got all the settings dialed in except the smoothing.

noiseremoval.JPG
mic_on_ultrapod.JPG
 
Best way to remove noise: prevention. How are you getting audible hiss in your recording in the first place? Fix it at the source and you won't have to go through these noise reduction contortions.
 
bouldersoundguy said:
Best way to remove noise: prevention. How are you getting audible hiss in your recording in the first place? Fix it at the source and you won't have to go through these noise reduction contortions.
Click to expand...

I don't think it's avoidable for what I'm doing. I'm recording people from about 3 feet away with the DR-40 set to 70 input level. From only a few inches away yes I'm sure the signal to noise ratio would be much better.

Maybe I need a shotgun mic...
 
winkosmosis said:
I don't think it's avoidable for what I'm doing. I'm recording people from about 3 feet away with the DR-40 set to 70 input level. From only a few inches away yes I'm sure the signal to noise ratio would be much better.

Maybe I need a shotgun mic...
Click to expand...

More likely radio tie tack mics!

This is yet another case of using the wrong or inadequate gear for a purpose for which it really was not intended.

The Tassy 144 is a good, very useful interface but the mic amps are far from the quietest on the market.
Same for the C1, never used one but again, a fair starter mic..IF you are using it at a foot (300mm) or so, past that and its self noise will start to intrude*.

There is a wealth of PC recording kit about now that is perfectly suited to the bedroom jockey who wants to yodel at a mic and/or plug in a guitar or mic an amp but such gear soon shows its limitations if you attempt to use it even a tiny bit outside its "comfort| zone".

Bottom line, sorry to be blunt. You Can't Do This On the Cheap! Not very well,anyway.

Last thing. Don't use 96kHz. There might just be a reason for this with some esoteric plugins but I doubt you have those. Run at 44.1Khz and 24 bits. In any event! When you have noise issues it is standard practice to LIMIT bandwidth!

*Of course! The Bellringer mic MIGHT have a noise figure to rival a Rode but somehow I doubt it!
Dave.
 
ecc83 said:
More likely radio tie tack mics!

This is yet another case of using the wrong or inadequate gear for a purpose for which it really was not intended.

The Tassy 144 is a good, very useful interface but the mic amps are far from the quietest on the market.
Same for the C1, never used one but again, a fair starter mic..IF you are using it at a foot (300mm) or so, past that and its self noise will start to intrude*.

There is a wealth of PC recording kit about now that is perfectly suited to the bedroom jockey who wants to yodel at a mic and/or plug in a guitar or mic an amp but such gear soon shows its limitations if you attempt to use it even a tiny bit outside its "comfort| zone".

Bottom line, sorry to be blunt. You Can't Do This On the Cheap! Not very well,anyway.

Last thing. Don't use 96kHz. There might just be a reason for this with some esoteric plugins but I doubt you have those. Run at 44.1Khz and 24 bits. In any event! When you have noise issues it is standard practice to LIMIT bandwidth!

*Of course! The Bellringer mic MIGHT have a noise figure to rival a Rode but somehow I doubt it!
Dave.
Click to expand...

I'm recording with the DR-40 not the US-144, at people's houses etc. Not sure how the noise on XLR inputs compares to similar devices but the major issue I've heard of is "helicopter sounds" which I haven't experienced.

I'm using 96khz for 2 reasons: Capture as much information as possible for posterity; and so I can downsample to both 44.1khz for CD and 48khz for video. I know a lot of people record in 48khz and downsample to 44.1khz but of course the interpolation from such a close sample rate necessarily reduces quality more than from a higher sample rate. It might not be noticeable but I'm OCD and I just don't want to downsample 48 to 44.1, much less record at 44.1 and upsample to 48!!

What about an entry level shotgun mic like this Rode NTG-1? http://www.bhphotovideo.com/c/product/367746-REG/Rode_NTG_1_NTG_1_Condenser_Shotgun_Microphone.html
 
you will not be losing anything for posterity by sampling at 44 or 48 kHz. Very few mics get to 20kHz and even if they did, what is up there that is useful or audible?

I can understand a producer using 96, even 192k if he was recording the Once a Century performance of 6 nation's massed choirs and three orchestras performing Stella Squeakygate's Symphony #1 conducted by the legendry Fartfinger who has come out of retirement just for the event...But a Behringer mic and a middle range portable recorder? Don't want to be rude but come on!

I would have thought the video MP4 etc compression would fork the quality far more than any sampling sheenanagins!

But all the best luck. There are some video guys here who I am sure will chip in.

Dave.
 
If you're OCD about quality then find a way to record clean audio. It's kind of self defeating to use 96k to record noisy audio.
 
ecc83 said:
you will not be losing anything for posterity by sampling at 44 or 48 kHz. Very few mics get to 20kHz and even if they did, what is up there that is useful or audible?

I can understand a producer using 96, even 192k if he was recording the Once a Century performance of 6 nation's massed choirs and three orchestras performing Stella Squeakygate's Symphony #1 conducted by the legendry Fartfinger who has come out of retirement just for the event...But a Behringer mic and a middle range portable recorder? Don't want to be rude but come on!

I would have thought the video MP4 etc compression would fork the quality far more than any sampling sheenanagins!

But all the best luck. There are some video guys here who I am sure will chip in.

Dave.
Click to expand...

I think in general there are a lot of misconceptions about how digital sound recording works. You mentioned that microphones generally can't pick up over 20khz hence sampling more than 44.1khz is useless. But the Nyquist theorem only applies here if you're talking about continuous tones. The sounds we're recording aren't continous sine waves, they're irregular and erratic. If you spend time on the ocean you know that waves come in different shapes and even have waves on top of them and crossing them... So the fact that the mic has little response over 20khz is irrelevant-- We're talking about sampling waves of all forms, of all frequencies, and usually not in phase with the sample rate, in order to recreate their shapes. A wave contains infinite information-- It's not encoded in MIDI. Why not sample as much of it as we can? There's a 1967 recording of Nelson Mandela on a Dictaphone that was just released. You can tell it's Mandela's voice and you can understand what he's saying. Does that mean there would be no value in better quality recording? It's like shooting in RAW vs JPEG. I'm a photographer and I shoot RAW only. Some people use JPEG because they can't tell the difference. But you only get one chance to take a photo or make a recording. I believe in doing it the best you can with the equipment you have. Hard drive space is cheap, processing power is cheap. Heck, the human voice tops out around 4000hz so does that mean I should sample at 8khz like a cell phone?

Here is an article I found that is all about what actually happens when sound is sampled. Much more helpful for understanding the process than the "twice the frequency" thing-- http://www.wescottdesign.com/articles/Sampling/sampling.pdf

Whether you or I have brains capable of distinguishing 44.1khz vs 96khz is another story. But I know I can tell when audio has been resampled between 44.1khz and 48khz. It's like resizing a photo, you'll always end up with less sharpness. But if you downsize from a larger resolution, the results are better.
 
Last edited:
"I think in general there are a lot of misconceptions about how digital sound recording works."

Indeed there are Winkosmosis and YOU are pedalling some of them. I am certainly no kind of expert on digital sound but then I can tell from your responses that neither are you. If you wish to stuff your hard drives and stress your processors with higher sampling rates that is your prerogative but do not promote misinformation in HR especially in the newbies section.

You "quote" the Nyquist theory but conveniently omit Fourier which tells us that ANY waveform can be broken down into sines and so long as you have enough of them WITHN THE PASSBAND OF THE SYSTEM. the waveform can be accurately described. For our purposes that "passband" is the limit of human hearing. So no, we should not restrict a recording channel to 4kHz because that is the limit of the human voice because there ARE frequencies above that (NOT that we need them for voice recognition. Heard of the telephone?) .

No waveform, however complex, contains an infinite amount of information. You can only have "information" if you have changes of state and they require energy and we don't have nearly enough of that leave alone an infinite amount!

These sorts of arguments have been trotted out since the dawn of "hi fi" . A similar one was promulgated about transient response which suggested that you needed systems with a response several time the range of human hearing. Cobblers, the transient response of a system is entirely described by its passband and damping. The passband once again for us is 20k (your dog/cat/bat will not agree but then some animals see into the ultraviolet. Do WE need that coming out of our tellies?) .

Digital recording, even 16bit 44.1kHz is the best general purpose recording system we have ever had. It knocks chunks off tape for flatness of frequency response, width of same, noise level and distortion (not to mention a complete absence of W&F, dropouts and scrape flutter sidebands) Digital WILL reconstruct waveforms. Try recording a 1000Hz squarewave on a Studer!

As I say, I am no digital guru but I have read widely in the last 8 years or so and seen all the myths demolished, e.g. better "resolution" of 24 bits or higher sampling rates. There is even a very big school of thought that says sampling past 48kHz is LESS accurate because real world converters are optimized for the lower rates!

So, go stuff your own drives if you wish. I am saying to noobs, "you don't need past 48kHz sampling. And certainly not if all you have is Audacity and a Behringer microphone!"

Oh! And don't trot out the Photographic analogue. It is not a good one.

Dave.
 
ecc83 said:
"I think in general there are a lot of misconceptions about how digital sound recording works."

Indeed there are Winkosmosis and YOU are pedalling some of them. I am certainly no kind of expert on digital sound but then I can tell from your responses that neither are you. If you wish to stuff your hard drives and stress your processors with higher sampling rates that is your prerogative but do not promote misinformation in HR especially in the newbies section.

You "quote" the Nyquist theory but conveniently omit Fourier which tells us that ANY waveform can be broken down into sines and so long as you have enough of them WITHN THE PASSBAND OF THE SYSTEM. the waveform can be accurately described. For our purposes that "passband" is the limit of human hearing. So no, we should not restrict a recording channel to 4kHz because that is the limit of the human voice because there ARE frequencies above that (NOT that we need them for voice recognition. Heard of the telephone?) .

No waveform, however complex, contains an infinite amount of information. You can only have "information" if you have changes of state and they require energy and we don't have nearly enough of that leave alone an infinite amount!

These sorts of arguments have been trotted out since the dawn of "hi fi" . A similar one was promulgated about transient response which suggested that you needed systems with a response several time the range of human hearing. Cobblers, the transient response of a system is entirely described by its passband and damping. The passband once again for us is 20k (your dog/cat/bat will not agree but then some animals see into the ultraviolet. Do WE need that coming out of our tellies?) .

Digital recording, even 16bit 44.1kHz is the best general purpose recording system we have ever had. It knocks chunks off tape for flatness of frequency response, width of same, noise level and distortion (not to mention a complete absence of W&F, dropouts and scrape flutter sidebands) Digital WILL reconstruct waveforms. Try recording a 1000Hz squarewave on a Studer!

As I say, I am no digital guru but I have read widely in the last 8 years or so and seen all the myths demolished, e.g. better "resolution" of 24 bits or higher sampling rates. There is even a very big school of thought that says sampling past 48kHz is LESS accurate because real world converters are optimized for the lower rates!

So, go stuff your own drives if you wish. I am saying to noobs, "you don't need past 48kHz sampling. And certainly not if all you have is Audacity and a Behringer microphone!"

Oh! And don't trot out the Photographic analogue. It is not a good one.

Dave.
Click to expand...


Sound can be modeled as sine waves, but sound isn't sine waves and they aren't sampled as number representing frequencies. The actual pressure variations are recorded. When you talk about reproducing a sound accurately with a sine wave, you're talking about a continuous tone. Real sound has variations from peak to valley of every single vibration. The more of that information you sample, the more accurately the sound is reproduced. You would be right IF all sounds were actually simple sine waves.

In any case, there is zero harm in recording as much information as possible. Hard drive space is cheap. I have more than enough processing power. Why are you worried about drive space and "stressing the CPU"? Do you think the CPU will wear out faster?

BTW if I'm reading your post right, you believe 48khz is useful over 44.1khz. I've seen many forum posts by anti-96khz folks that 48khz is noticeably better than 44.1khz. Why would it be better if your ear can't hear details beyond 1/20,000th of a second? Why wouldn't it be exactly the same?

Regardless, I need a sample rate that can be resampled properly to BOTH 44.1khz and 48khz. Downsampling from 48khz to 44.1khz requires that all 44,100 samples each second be interpolated from the existing 48,000 samples. The more samples you start with, the less the quality loss from interpolation. It's basically the same algorithms whether you're talking about photos or sound.

Here are some articles on sample rates:

http://www.soundstagehifi.com/index...-nyquist-and-appropriate-sampling-frequencies
 
Last edited:
winkosmosis said:
I'm recording with the DR-40
Click to expand...

You might have fun no matter what way you approach this.
I have a DR40 and find it to be pretty noisy.
It's certainly too noisy to consider using it for 'proper' recordings. I only use it for the occasional youtube video.

To be honest, I was pretty disappointed by it.
 
Steenamaroo said:
You might have fun no matter what way you approach this.
I have a DR40 and find it to be pretty noisy.
It's certainly too noisy to consider using it for 'proper' recordings. I only use it for the occasional youtube video.

To be honest, I was pretty disappointed by it.
Click to expand...

In that case I guess I should stick with the hardware I have and focus on noise removal, as I was planning. Otherwise I'd have to spend a lot of money on a better mic and PCM.

So how exactly do I figure out what frequency smoothing setting to use?
 
By looking at your sample in something that has an EQ plug in / display, setting a tight Q, maximum EQ gain, fair bit of volume, listening through headphones and sweeping up and down until you find the frequency at which you rip the headphones off your head and swear at your wife...
 
Q Should I use high sample rates?

That^ shows that except in exceptional circumstances there is nothing to be gained by going past 48kHz (which is needed for video otherwise 44.1kHz) .

I don't mind, as I said, if you stuff your own drives I just want noobs to know that like so many "must do" myths about audio there is no need for higher sample rates.

Dave.
 
winkosmosis said:
In that case I guess I should stick with the hardware I have and focus on noise removal, as I was planning. Otherwise I'd have to spend a lot of money on a better mic and PCM.
Click to expand...

To insist on high sample rates for "better quality" and then settle for gear that makes so much noise you have to use noise reduction software is fundamentally irrational. It's no wonder there's a disconnect between the advice you want and the advice you're getting. The right answer is always going to be "fix it at the source".
 
bouldersoundguy said:
To insist on high sample rates for "better quality" and then settle for gear that makes so much noise you have to use noise reduction software is fundamentally irrational. It's no wonder there's a disconnect between the advice you want and the advice you're getting. The right answer is always going to be "fix it at the source".
Click to expand...

I agree. It's kind of weird, specially when the "better quality" in higher sample rates is illiusory, whereas the noise is not.
 
ecc83 said:
Q Should I*use high sample rates?

That^ shows that except in exceptional circumstances there is nothing to be gained by going past 48kHz (which is needed for video otherwise 44.1kHz) .

I don't mind, as I said, if you stuff your own drives I just want noobs to know that like so many "must do" myths about audio there is no need for higher sample rates.

Dave.
Click to expand...

I don't know how to explain this any better. I will not sample at 48khz. Among output format will be YouTube, CD, and MP3 which are 44.1khz. Assuming 48khz is good enough as you say, you're ignoring the fact that interpolating all 44100 samples per second from those 48000 will noticeably reduce quality. Hence sampling at 96khz. It's good enough to downsample both to 44.1 and 48.

Are you under the impression that downsampling 48khz to 44.1khz has no quality loss? Do you believe that the 3900khz are just extra data that is discarded?
 
Last edited:
gecko zzed said:
I agree. It's kind of weird, specially when the "better quality" in higher sample rates is illiusory, whereas the noise is not.
Click to expand...

That makes no sense. There is no harm in recording at 96khz, and even if you BELIEVE it's illusiary there is still a clear benefit for downsampling to 44.1 and 48. Just because I have entry level equipment doesn't mean I shouldn't sample the best I can. The more data you capture, the better for postprocessing, hence better for noise reduction etc. Especially since algorithms for reducing noise seem to be continually getting better.

What you're saying is equivalent to "You only have a $500 camera lens so you should shoot photographs in JPEG". Capturing more information is always better.

I'll never understand this obsession with limiting sampling to 44.1 or 48. What on earth do you guys have to gain by insisting on not sampling at 96khz? How on earth does it harm you?

BTW I emailed Tim Wescott, the author of the sampling article I posted above. Here is a part of one of his replies:

Be that as it may, in this case I think that there are sound technical
reasons for the naysayers to be on the wrong side of the argument.
First, it's not so much that your ear works as a brick-wall filter up to
20kHz exactly and no farther -- as the frequency goes up, your
sensitivity to sound diminishes, but doesn't go away between 19.999kHz
and 20.001kHz. Given that, then a sampling rate of 44.1kHz is awfully
low.

Moreover, some people can hear well above 20kHz (for instance, my
sisters, when they were younger, could hear "ultrasonic" burglar alarms
that operated at 25kHz or so -- and may still). For those people you're
definitely cutting out frequencies that they are sensitive to.

And here's something else I've found. Read the comments: http://recordinghacks.com/articles/the-world-beyond-20khz/

Serge
December 19th, 2013 at 8:54 am
As an EE and audio enthusiast I see an immediate and rather simple explanation that ties together much of what I have known so far:
Our ear detection of the continuous audio waves seems like limited to about 18 – 25 kHz. However short time bursts of waves, though not perceived as a ’sound’, still being processed by the brain as an important ‘additive’ of the pulse (or step) audio signal. Why? Okay, let’s remember that Fourier representation of a step signal is a multitude of wave forms of different frequencies and envelopes, phase aligned such, that rising parts of all of them coincide with the moment of the step. The step is always as ’sharp’ as high is the frequency of the most high frequency wave ingredient we are able to provide. And here is the kicker: Those higher frequency waves have very short envelopes, and thus, (back to our ears) though not detected in a continuous wave test, still contribute to the shape of the pulse and processed by our brains to give us impression of a natural sound, of course if we provide a full path from the source through amplification and speakers.

Wait, but there's more. The brain can detect timing differences as low as 5-15 microseconds which is well beyond each sample even at 96khz http://www.digitalprosound.com/Htm/SoapBox/soap2_Apogee.htm

There you have it. The ear doesn't have a sharp cutoff. The ear and brain don't just determine frequencies, they perceive the actual waveform. You can hear sounds shorter than 1/20,000th of a second. You can detect timing differences beyond even what 96khz captures. Sampling above 44.1khz therefore seems to be useful even if 44.1khz is "good enough". The presence of a small amount of noise doesn't mean you should sample at less than what is useful. Heck, if that were the case why not sample at 22khz? Should I record 22khz 64kbps mp3? How well will noise removal work on those files?

You only get to record once, and in many cases these will be the only recordings of these folks that still exist in a decade or two. There's no reason not to record as best I can with the equipment I have. Not to save drive space, not because people who don't really understand the physics and biology behind hearing swear up and down that the human ear has a brick wall filter that makes sounds shorter than 1/20,000th sec useless.
 
Last edited:
Getting back on topic, since no one has answered me yet-- Why do I get good results even without frequency smoothing? Is it necessary?

And is there a better way to remove noise than Audacity's noise removal tool?
 
The point is you are wasting any notional quality advantage through sampling at 96 by:

a) tracking with less than optimal quality; and
b) having to deal with the consequences of that by using a sub optimal process (noise reduction).
 
You must log in or register to reply here.
Back
Top