Question about KHz

  • Thread starter Thread starter the7azen
  • Start date Start date
T

the7azen

New member
Hello ..
I am going to buy a condernsire with 30Hz-20kHz frequency response
And a usb mic port pro with 44.1kHz and 48.0kHz

Does it match together ..
Does it make different if the 2 objects .. one have high KHs and the other have low one

and thx
 
No, they're two different things...

The microphone frequency response refers to the range of frequencies that the microphone can transduce.

The spec you've reference for the soundcard is the sampling rate... the frequency at which digital samples are taken.

The highest frequency that can be represented by a digital signal is half of the sampling rate (Nyquist theorem).
 
The two here are not entirely related. The first is the frequency response of the microphone - the frequencies which will be 'heard' and recorded.

The second number you have there is the supported sample rates of the audio interface. This is related to 'sound quality' - the higher the better. Depending on your PC specs you may want to record at 44.1Khz (CD quality) rather than 48Khz.

EDIT: Beaten to it - but there's your basics. Frequency response is fairly self explanatory, but you may wish to look into sound quality i.e. sample rate and bit depth.
 
Do yourself a BIG favor and stay away from usb mics and adapters.
They are basically made for PODCASTING (radio chat shows), not music recording.

The analog-to-digital converters in their mini-soundcard are mediocre for music.
The poor monitoring in them will give you fits trying to overdub new tracks.
The short cords will introduce a LOT of computer fan noise.

Get yourself a real audio interface. Here's a good guide:
http://www.tweakheadz.com/soundcards_and_audio_interfaces.htm

(you'll want to bookmark and read through ALL of Tweak's Guide...)
 
Of course it matters. But not really that much.

Human hearing is roughly 20Hz - 20kHz.

48kHz is how many samples are taken per second. Which may or may not reproduce frequencies outside of the range of human hearing.

We are still sensitive to those frequencies outside of 20Hz-20kHz by other means. And the specs for the mics range is what it is tested for, the microphone will still pick up frequencies outside that range, just not at a level worth noting on specs. Or at a frequency not tested for. Not that it matters if your cheap PC speakers only reproduce 50Hz-16kHz. But it does matter because the gear will TRY to reproduce those pitches and fail and otherwise not be functioning optimally. But it doesn't matter since that's not the frequencies that you're most interested in, in most cases.
 
Shadow_7 said:
Of course it matters. But not really that much.

Human hearing is roughly 20Hz - 20kHz.

48kHz is how many samples are taken per second. Which may or may not reproduce frequencies outside of the range of human hearing.

We are still sensitive to those frequencies outside of 20Hz-20kHz by other means. And the specs for the mics range is what it is tested for, the microphone will still pick up frequencies outside that range, just not at a level worth noting on specs. Or at a frequency not tested for. Not that it matters if your cheap PC speakers only reproduce 50Hz-16kHz. But it does matter because the gear will TRY to reproduce those pitches and fail and otherwise not be functioning optimally. But it doesn't matter since that's not the frequencies that you're most interested in, in most cases.
Click to expand...

Sorry if I've brought this up ad nauseum.

According to Nyquist theory, the sampling rate only needs to be slightly above double the highest audible frequency to reproduce the wave correctly.

44.1Khz is really all you need.
 
Unless you're recording the 50kHz vocalizations of bats for scientific research. So yes and no.
 
Shadow_7 said:
Unless you're recording the 50kHz vocalizations of bats for scientific research. So yes and no.
Click to expand...

Um, no. The sample rate is set to catch the upper and lower portions of the waveform at any given point in time and hopefully create the same waveform digitally. Therefore to sample 20khz you need to take two samples per one cycle in order to be able to catch both peaks of the wave: 40Khz or above. In order to get a sample at 50khz you'd need to have a sample rate above 100khz.
 
PhilGood said:
Um, no. The sample rate is set to catch the upper and lower portions of the waveform at any given point in time and hopefully create the same waveform digitally. Therefore to sample 20khz you need to take two samples per one cycle in order to be able to catch both peaks of the wave: 40Khz or above. In order to get a sample at 50khz you'd need to have a sample rate above 100khz.
Click to expand...

Um duhhhh. That's what I meant. i.e. 44.1kHz is NOT all you need. Depending on the purpose anyway.
 
Shadow_7 said:
Um duhhhh. That's what I meant. i.e. 44.1kHz is NOT all you need. Depending on the purpose anyway.
Click to expand...

Well, I see that we were on the same page, however the other position I was making is on how some people try to sell the snake oil that somehow it affects the accuracy of the high end and tell you that 96Khz is better, which for audio is untrue. That's what I meant by 44.1Khz is all you need.
 
PhilGood said:
Well, I see that we were on the same page, however the other position I was making is on how some people try to sell the snake oil that somehow it affects the accuracy of the high end and tell you that 96Khz is better, which for audio is untrue. That's what I meant by 44.1Khz is all you need.
Click to expand...

The required steepness of the anti-aliasing and anti-imaging filters will cause some passband attenuation using a 44.1kHz sample rate. This is not snake oil as it is easily verifiable with D/A/D roundtrip of a test signal. It is mostly ameliorated at 48kHz but that depends on the quality of the filter implementation. Some converters (mostly older ones) perform worse than others. It is completely gone at 20kHz with any reasonably modern converter at an 88.1kHz sample rate.

Another consideration is the need for oversampling in processing. Arguably it is better to record at a higher sample rate and use processing that thus does not use oversampling (or uses a minimal amount) rather than having every plug in a chain running its own upsampling and downsampling routines. Beyond the obvious waste of CPU cycles, the effect will be similar to multiple D/A/D roundtrip in terms of passband attenuation. The undesirable alternative is aliasing, which is unfortunately common in several popular processing applications . . .
 
PhilGood said:
Well, I see that we were on the same page, however the other position I was making is on how some people try to sell the snake oil that somehow it affects the accuracy of the high end and tell you that 96Khz is better, which for audio is untrue. That's what I meant by 44.1Khz is all you need.
Click to expand...

Snake oil or not, 96kHz has it's uses if you need multiple outputs like 48kHz for video and 44.1kHz for CDs.

Or if you just want to run your 96kHz soundcard at specifications. Assuming all of the other gear/parts needed to reproduce that specification. At a minimum it can make the pets more interesting when doing so. i.e. The cat stalking the studio monitors trying to find that bird in the speaker and stuff like that.
 
Also, higher sampling rates allow better precision for pitch detection and scaling algorithms based on FFTs. With twice the sampling rate, if you keep the length (in milliseconds) of the sample buffer the same, this means you have twice as many samples in the sample buffer, which translates to twice as many FFT buckets, and thus twice the initial pitch detection precision (before you start trying to throw maxima finding algorithms at it). This can result in an audible difference in the resulting sound.
 
According the Yamaha there are harmonics above and below the audible spectrum. Their natural sound amps go 10hz to 100khz. I've got one, and don't hear a bit of difference between it and a Sony that does 20hz to 20khz. I've got one of those too.
 
buzzard bass said:
According the Yamaha there are harmonics above and below the audible spectrum. Their natural sound amps go 10hz to 100khz. I've got one, and don't hear a bit of difference between it and a Sony that does 20hz to 20khz. I've got one of those too.
Click to expand...

If your DAC only does 48kHz, or is only set to 48kHz (or less), it's not going to send 96kHz down the line to those speakers. Lots of parts to make the whole.

If your mic and/or preamp and/or editing software and/or media player brickwalls at 20Hz-20kHz, you wouldn't hear a lick of difference. That and you have to have the physical ability to hear that range. Or sense it by other means. A 44100Hz CD is only going to reproduce 22050Hz pitches. Not to imply that there are any 22050Hz pitches on said CD. So even if there is a difference you're not likely to notice. The difference is very minor, but it could be the difference between, that's nice, and what's missing?


In linux with sox, or at least part of it.

$ play -n synth <duration> sin <frequency> gain -0.5

$ play -n synth 1.0 sin 17000 gain -0.5
(produces 1 second of a sin wave at 17kHz at a reasonable volume.)
 
dgatwood said:
Also, higher sampling rates allow better precision for pitch detection and scaling algorithms based on FFTs. With twice the sampling rate, if you keep the length (in milliseconds) of the sample buffer the same, this means you have twice as many samples in the sample buffer, which translates to twice as many FFT buckets, and thus twice the initial pitch detection precision (before you start trying to throw maxima finding algorithms at it). This can result in an audible difference in the resulting sound.
Click to expand...
There might be more buckets, but the number of waves would be the same. I could be wrong, but I would assume that it takes a certain number of cycles at a certain frequency for something to determine the pitch (or frequency) of the sound. No matter how many sample you have of that sound, it will still be the same number of cycles long.
 
PhilGood said:
Sorry if I've brought this up ad nauseum.

According to Nyquist theory, the sampling rate only needs to be slightly above double the highest audible frequency to reproduce the wave correctly.

44.1Khz is really all you need.
Click to expand...
technically but the 'brick wall' filters that are needed to prevent aliasing cause issues of their own.
A higher sampling rate is better when available.
 
Farview said:
There might be more buckets, but the number of waves would be the same. I could be wrong, but I would assume that it takes a certain number of cycles at a certain frequency for something to determine the pitch (or frequency) of the sound. No matter how many sample you have of that sound, it will still be the same number of cycles long.
Click to expand...

There is a certain minimum time period involved. That is why you can't reduce the latency usefully below a certain point---the buffer has to hold a reasonable portion of the wave. That's also why they provide a separate pitch correction algorithm for bass instruments. That's not what I'm talking about, though. You have to start with the assumption that you have enough of a wave for the fundamental to be the strongest FFT bucket anyway or pitch correction will flat out fail.... Well, more accurately, it will fluctuate wildly, and a properly behaving pitch correction algorithm will treat it as being random noise rather than a correctable pitch....

In an FFT, you transform a signal from the temporal domain to the frequency domain. So instead of expressing the signal as a series of voltages at discrete points in time, you are expressing the same values as the sum of sine waves at various frequencies at various volumes (positive or negative). You can then determine the fundamental frequency by looking for the largest value (basically) and seeing which bucket it is in, but that only gets you the frequency accurate to +/- the width of the bucket, which is determined by the number of buckets.

To get the real pitch, you have to take the largest value and the value to either side, then apply curve matching to find the exact peak. That peak might be... say 7/8ths the way between two adjacent buckets. If you start out with twice the accuracy, you can now find the peak with greater accuracy because you started with twice as many values in the frequency domain.

Incidentally, this also applies to algorithms that do pitch detection in the time domain. The way you do that is to (usually filter the heck out of the signal and then) try to count zero crossings or peaks and try to average out those values in a sensible way. There are many algorithms for doing this, and in every case, you need to calculate the crossing or the peak as accurately as possible, which means you get better accuracy with more points.
 
Lt. Bob said:
technically but the 'brick wall' filters that are needed to prevent aliasing cause issues of their own.
A higher sampling rate is better when available.
Click to expand...

Basically. The bandpass filtering still has to occur when you downsample to the final output format (assuming that you aren't releasing your recording in a 96 kHz format or whatever). That said, an analog bandpass filter cannot realistically perform as well as a well-written digital bandpass filter, so there's a very definite win there, mostly in terms of phase accuracy.

http://www.mstarlabs.com/dsp/antialiasing/antial.html
http://en.wikipedia.org/wiki/Sinc_filter

I hesitate to call it a brick-wall filter because it's technically just an approximation of one, but....
 
You must log in or register to reply here.
Back
Top