Noise Removal in Audacity - Frequency Smoothing

[winkosmosis]

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).

I don't know what's so difficult to understand about capturing as much information as possible, ESPECIALLY for future noise reduction and postprocessing. I also don't know what's so difficult to understand about sampling at 96khz for optimal quality when downsampling to 44.1 and 48 (or keeping 96 for DVD). Stop worrying about my hard drive space. Just accept that I'm using 96khz and provide useful advice for noise reduction if you have any.

 

[ecc83]

I don't know what's so difficult to understand about capturing as much information as possible, ESPECIALLY for future noise reduction and postprocessing. I also don't know what's so difficult to understand about sampling at 96khz for optimal quality when downsampling to 44.1 and 48 (or keeping 96 for DVD). Stop worrying about my hard drive space. Just accept that I'm using 96khz and provide useful advice for noise reduction if you have any.

THAT is the myth! You are NOT capturing any more information! The mic simply does not fetch it in!
Then there IS some evidence that most AI are not optimized for 96k+ and so you might be doing more damage that way.

And BTW, the ear DOES have a brickwall cutoff. From what is known if you "don't have the hairs for it, you cannot hear it".

And PLEASE read things aright? I have said, I do not care about YOUR HD space, just don't pedal the idea to the rest of the people here.

Dave.

 

[Mr Clean]

and provide useful advice for noise reduction if you have any.

Get better equipment for your requirements or use better positioned microphones so that you don't record hiss and noise in the first place. After much fannying about trying to remove noise, you'll probably want to re-record the audio again. You're pretty much pissing in the wind if you think you're going to remove noise and hiss and have anything half usable, especially for voice recording.

Try the C1 on a boom stand above the people/speakers you're recording. Closer to the source + less gain = better recording and less messing about after the fact.

I have to admit that I'm a little confused that you state that you're OCD about getting things as good as they can be. Recording at 96khz, photogragraphing raw, etc. But you're using inferior tools to achieve high quality recordings? I don't think you're OCD enough.

 

[ecc83]

Attached is a frequency chart of some measurement microphones (gen purp music mics either don't give a graph, cut off too soon or tell porkies) .

You can see that most of them get to around 20kHz then FALL OFF A CLIFF! Nothing, nada, buggerall past 20k. And this is because the HF response of a mic is maintained up to about the point where the circumference of the diaphragm is equal to a wavelength and (best Cilla voice, on topic atmo) surprise, surprise, that turns out to be about 22,000 Hz for a 1/2" diaphragm. If an ultrasonic pressure wave cannot generate a net force in the mic D there can ne no movement and no lekky!

Now IF the AI actually converts frequencies up to about 48kHz (and many don't you know, they leave the filters set for 44.1kHz! 96kHz capability just looks good on the box!) you are not converting anything from the mic just residual electronic noise.

NB the couple that get to ~120kHz are 1/4" D types and have very low sensitivities.

And there is for once a good "light analogy". Why do we need electron microscopes?
Dave.

 

[winkosmosis]

Get better equipment for your requirements or use better positioned microphones so that you don't record hiss and noise in the first place. After much fannying about trying to remove noise, you'll probably want to re-record the audio again. You're pretty much pissing in the wind if you think you're going to remove noise and hiss and have anything half usable, especially for voice recording.

Try the C1 on a boom stand above the people/speakers you're recording. Closer to the source + less gain = better recording and less messing about after the fact.

I have to admit that I'm a little confused that you state that you're OCD about getting things as good as they can be. Recording at 96khz, photogragraphing raw, etc. But you're using inferior tools to achieve high quality recordings? I don't think you're OCD enough.

I can't afford $1000 worth of mics and a better PCM recorder at the moment. I can afford to record at the best quality possible because it costs nothing! If I record at 48khz and resample to 44.1khz for CD, or vice versa, I'll be significantly degrading quality. That just makes zero sense to me. Why make the sound worse than it has to be?

Even at 1ft away with a normal person's speaking voice there's noise with the C1. I looked into the $250 Rode NTG-1 shotgun mic but it has a relatively low SNR, only 76dB. I have a friend in the film industry who uses the NTG-1 with a DR-40 and he says he has to do noise reduction in post. I'm thinking about buying the $200 Rode NT1-A as it's supposed to be the "world's quietest studio mic"

 

[winkosmosis]

Attached is a frequency chart of some measurement microphones (gen purp music mics either don't give a graph, cut off too soon or tell porkies) .

You can see that most of them get to around 20kHz then FALL OFF A CLIFF! Nothing, nada, buggerall past 20k. And this is because the HF response of a mic is maintained up to about the point where the circumference of the diaphragm is equal to a wavelength and (best Cilla voice, on topic atmo) surprise, surprise, that turns out to be about 22,000 Hz for a 1/2" diaphragm. If an ultrasonic pressure wave cannot generate a net force in the mic D there can ne no movement and no lekky!

Now IF the AI actually converts frequencies up to about 48kHz (and many don't you know, they leave the filters set for 44.1kHz! 96kHz capability just looks good on the box!) you are not converting anything from the mic just residual electronic noise.

NB the couple that get to ~120kHz are 1/4" D types and have very low sensitivities.

And there is for once a good "light analogy". Why do we need electron microscopes?
Dave.

That's for continuous tones--- But the value of the higher sample rate isn't in capturing continuous tones, but capturing a more accurate model of the waveform--- Is the upper frequency limit of the mic the same as the shortest length of sound it can capture? I don't think so. If you make a click that's 1/96000th of a second, a mic will pick it up


When I get home I'm going to try this to visualize what waveforms look like from my mics and I will post screenshots. I'm betting we won't be seeing sine waves but complex waveforms with details shorter than 1/20000th sec http://www.sonicvisualiser.org/

 

[ecc83]

That's for continuous tones--- But the value of the higher sample rate isn't in capturing continuous tones, but capturing a more accurate model of the waveform--- Is the upper frequency limit of the mic the same as the shortest length of sound it can capture? I don't think so. If you make a click that's 1/96000th of a second, a mic will pick it up


When I get home I'm going to try this to visualize what waveforms look like from my mics and I will post screenshots. I'm betting we won't be seeing sine waves but complex waveforms with details shorter than 1/20000th sec Sonic Visualiser

You are just going round in mad, inattentive, mythical circles. Have you never examined sounds on an oscilloscope?

Dave.

 

[winkosmosis]

You are just going round in mad, inattentive, mythical circles. Have you never examined sounds on an oscilloscope?

Dave.

No but I have never seen a perfect sine wave on an oscilloscope and I have never seen a perfect sine wave on the ocean

 

[ecc83]

No but I have never seen a perfect sine wave on an oscilloscope and I have never seen a perfect sine wave on the ocean

Well then (ignoring your misdirection and lack of understanding) sound waves resolve into more and more SINE waves as you increase the timebase frequency.
Why don't you let go if all this mid 70s "Hi Fi Choice" crap and learn how real systems work?

Once a sensing device becomes comparable to the wavelength of the signal out put drops to zero. Works for microphones, works for tape heads.

Dave.

 

[DaleVO]

I can't afford $1000 worth of mics and a better PCM recorder at the moment. I can afford to record at the best quality possible because it costs nothing! If I record at 48khz and resample to 44.1khz for CD, or vice versa, I'll be significantly degrading quality. That just makes zero sense to me. Why make the sound worse than it has to be?"

I found this interesting Xiph.org explanation, once upon a time, when I was trying to wrap my mind around sampling rates and bit depth (from 00:09:04 - 00:15:20 or so): http://m.youtube.com/watch?v=Ny7krNFAD1s

A little deeper discussion that let me put on my inner-geek: http://m.youtube.com/watch?v=d7kJdFGH-WI

 

[winkosmosis]

Here is a comparison of the 96khz 24 bit file and one resampled to 48khz 16 bit. You can see that the mic is picking up detail beyond 1/48000th of a second even though it's rated at 20hz-20khz. And you can see a big difference in the valleys of each "big" wave. Note that the dominant wave isn't a sine. So if you try to the voice (a 74 year old man's) based on Nyquist frequency, you'll end up with a very different sound.

Unfortunately I don't have the gear to split a mic signal and sample at two sample rates simultaneously

 

[ecc83]

DAW software only give a "representation" of a waveform or a spectrum,a piece of equipment that would be marked in a lab as "Not Calibrated: For Indication Only"

They are not £40,000 Real Time Analysers.

Why can you not accept the laws of physics? 1/2 inch mics S.T.O.P. at about 20kHz.

You cannot listen to bats with a U87 no matter that it is a fabulous mic and I want one!

96kHz is NOT getting any more information from a microphone past ~20k, COS IT IS FEKKIN' DEAF up there!

Dave.

 

[winkosmosis]

DAW software only give a "representation" of a waveform or a spectrum,a piece of equipment that would be marked in a lab as "Not Calibrated: For Indication Only"

They are not £40,000 Real Time Analysers.

Why can you not accept the laws of physics? 1/2 inch mics S.T.O.P. at about 20kHz.

You cannot listen to bats with a U87 no matter that it is a fabulous mic and I want one!

96kHz is NOT getting any more information from a microphone past ~20k, COS IT IS FEKKIN' DEAF up there!

Dave.

Even if the mic were literally not picking up any frequency higher than 20khz OR any sound shorter than 1/20000th of a second, you still have the issue that you're digitally sampling an analog signal that doesn't know or care what the phase of the sampling equipment is. Never will you get a wave that has crests and troughs which coincide exactly with the sampling points. So the more you have the more closely you can capture the real wave

 

[ecc83]

Even if the mic were literally not picking up any frequency higher than 20khz OR any sound shorter than 1/20000th of a second, you still have the issue that you're digitally sampling an analog signal that doesn't know or care what the phase of the sampling equipment is. Never will you get a wave that has crests and troughs which coincide exactly with the sampling points. So the more you have the more closely you can capture the real wave

Oh FFS! There is NOTHING THERE past 20kHz TO sample!!!



Dave.

 

[winkosmosis]

I found a good test subject... Insects. There are bugs here that chirp all night, similar to crickets but I don't think they're crickets. They make the same sound over and over so you can get sort of consistent recordings.

I resampled from 96khz and 48khz up to 384khz which is as close I can get to making a visual representation of an analog wave. I got some screenshots from parts of the recording that look similar in amplitude and overall frequency-- The song of the bugs seems to undulate.

1 second of audio:

Zoomed samples:

See the little waves picked up by the mic? I don't think those are noise because they're consistent and there are repeating patterns

 

[Mr Clean]

If I record at 48khz and resample to 44.1khz for CD, or vice versa, I'll be significantly degrading quality. That just makes zero sense to me. Why make the sound worse than it has to be?

I don't care for that argument. Knock yourself out recording at whatever sample rate you choose.

Even at 1ft away with a normal person's speaking voice there's noise with the C1.

That's because it's a rubbish $50 mic with probably less than $5 worth of electrics. Close mic a loud source like an acoustic guitar, vocals, and you could easily get away with using this mic. From a distance recording conversations, not so much. It's just not suited for that task.

I have a friend in the film industry who uses the NTG-1 with a DR-40 and he says he has to do noise reduction in post.

Ah that old chestnut! Why don't you ask him then? Ask him what tools he uses to remove noises and hiss and the settings he uses, etc? I'm willing to bet he isn't trying to rescue audio with Audacity.

You've argued every bit of advice you have received here so I'm guessing no matter what you're told you will refuse to take on board.

I wish you luck.

 

[bouldersoundguy]

Even if the mic were literally not picking up any frequency higher than 20khz OR any sound shorter than 1/20000th of a second, you still have the issue that you're digitally sampling an analog signal that doesn't know or care what the phase of the sampling equipment is. Never will you get a wave that has crests and troughs which coincide exactly with the sampling points. So the more you have the more closely you can capture the real wave

The problem with your concept of digital recording is that you seem to be unaware of reconstruction filters which more than adequately interpolate the shape of the wave between the sample points.

The main advantage of high sample rates was that less steep analog low pass filters could be used, but since practically all converters are of the oversampling type they also have that advantage even if the final output is 44.1 or 48. Even 96k digital audio is typically low passed at 20kHz.

I believe noisy audio -> 96k conversion -> noise reduction -> resampling to 44.1 and 48 will be of lower quality than clean audio -> 48k conversion -> resampling to 44.1. The noise and noise reduction do more damage than 96k can fix.

 

[winkosmosis]

The problem with your concept of digital recording is that you seem to be unaware of reconstruction filters which more than adequately interpolate the shape of the wave between the sample points.

The main advantage of high sample rates was that less steep analog low pass filters could be used, but since practically all converters are of the oversampling type they also have that advantage even if the final output is 44.1 or 48. Even 96k digital audio is typically low passed at 20kHz.

I believe noisy audio -> 96k conversion -> noise reduction -> resampling to 44.1 and 48 will be of lower quality than clean audio -> 48k conversion -> resampling to 44.1. The noise and noise reduction do more damage than 96k can fix.

When you say perfectly interpolated what you mean is "perfectly interpolated if the original waves were perfect sines"-- I stand by my statement that the more samples you have the more accurately the waveform gets reconstructed and interpolated

Why do you think it's better to sample at 48khz and resample to 44khz than to sample at 96khz and resample to 44khz? What quality advantage would your preferred process have?

 

[winkosmosis]

I don't care for that argument. Knock yourself out recording at whatever sample rate you choose.



That's because it's a rubbish $50 mic with probably less than $5 worth of electrics. Close mic a loud source like an acoustic guitar, vocals, and you could easily get away with using this mic. From a distance recording conversations, not so much. It's just not suited for that task.



Ah that old chestnut! Why don't you ask him then? Ask him what tools he uses to remove noises and hiss and the settings he uses, etc? I'm willing to bet he isn't trying to rescue audio with Audacity.

You've argued every bit of advice you have received here so I'm guessing no matter what you're told you will refuse to take on board.

I wish you luck.

The only advice I'm arguing is the advice to "use 48khz because your quality sucks anyway". I've been TRYING to explain that less than optimal quality in part of the process doesn't somehow mean that sampling at a lower rate somehow makes sense, and that the opposite makes more sense-- Sample at 96khz to capture more information for postprocessing and for better downsampling. Unfortunately the "Nyquist says..." dogma is pretty hard to cut through. Try as I might, I can't convince you that downsampling 48khz->44.1khz is worse than 96khz->44.1khz because you're convinced that 48 and 44.1 are both lossless reproductions of real sound because "Nyquist says..."

My friend uses Final Cut, which I can't since I'm using Windows. I'm still deciding which video editing program to use, which I'm sure will have its own noise reduction setup.

BTW I ordered the Rode NT1-A condenser, supposed to be the "quietest studio mic in the world". I'll report back on the noise level with the DR-40

 

[bouldersoundguy]

When you say perfectly interpolated what you mean is "perfectly interpolated if the original waves were perfect sines"-- I stand by my statement that the more samples you have the more accurately the waveform gets reconstructed and interpolated

I didn't say "perfectly", I said "more than adequately". The original waveform is essentially a series of sine waves.

Why do you think it's better to sample at 48khz and resample to 44khz than to sample at 96khz and resample to 44khz? What quality advantage would your preferred process have?

Again you're misrepresenting what I said. It's the whole signal chain I'm talking about, with the noise and the noise reduction. The advantage is that it wouldn't start out with audible noise and it wouldn't be subjected to noise reduction processing, which would more than for the extremely small advantage of 96k.