mikeash.com pyblog/friday-qa-2012-10-12-obtaining-and-interpreting-audio-data.html comments

Andrew Benson - 2019-08-01 17:54:28

Thu, 01 Aug 2019 17:54:28 GMT

Great article and still very relevant today. Thanks Mike!

A few people asked about performing FFTs. There's a great document from Apple with step-by-step instructions and code for doing exactly that using the Accelerate framework.

https://developer.apple.com/documentation/accelerate/vdsp/fast_fourier_transforms/finding_the_component_frequencies_in_a_composite_sine_wave

dougg3 - 2014-11-02 16:33:29

Sun, 02 Nov 2014 16:33:29 GMT

Thanks--your discussion on decibels was really helpful! It's the morning, so I may be full of it right now (correct me if I'm misinterpreting something), but 20*log10f(averageAmplitude) is not equal to 10*log10f(averageSquaredAmplitude) for the same reason that sqrtf(a*a + b*b) is not equal to a + b. For example, according to Google:

20*log10((10 + 20 + 30) / 3) = 26.021
10*log10((100 + 400 + 900) / 3) = 26.690

mikeash - 2012-10-21 01:25:55

Sun, 21 Oct 2012 01:25:55 GMT

For Nyquist to apply, you need to sample the signal for an infinite amount of time with perfect bandlimiting and unlimited precision. To reconstruct the signal exactly, you have to sum an infinite number of terms. Real-world systems must necessarily only approximate these conditions, and with approximations, the proof no longer applies. It would not surprise me if higher frequencies suffer from distortion under these real-world conditions.

jamie - 2012-10-19 22:23:30

Fri, 19 Oct 2012 22:23:30 GMT

Actually this isn't true. Nyquist-Shannon proves that a band limited signal will be reproduced *exactly*. Note that a DAC does not output a stepwise function, but each impulse is multiplied by the sinc function ( sin(x) / x ). See the wiki: http://en.wikipedia.org/wiki/Nyquist-Shannon_sampling_theorem#Reconstruction

Some other modulations do however pickup distortion at higher frequencies, particularly DSD or Sigma-Delta.

SSteve - 2012-10-19 00:03:31

Fri, 19 Oct 2012 00:03:31 GMT

When discussing sampling I think it's important to mention that as frequencies approach the Nyquist limit they are represented with more and more distortion. Just so people don't think it's 100% accurate all the way up. Personally, it doesn't make a lot of difference for me because I can't hear much above 13kHz. Loud drummer in my junior high school band.

Jake - 2012-10-16 22:07:13

Tue, 16 Oct 2012 22:07:13 GMT

Can I second that idea? I'd love to see how to do an FFT on raw audio data. If that's not being greedy :) Great article!

jamie - 2012-10-16 00:40:04

Tue, 16 Oct 2012 00:40:04 GMT

You might also do some theory on FFTs, how to window and analyze samples, and then use them for visualization or resynthesis.

Julian Iaccopucci - 2012-10-15 13:18:45

Mon, 15 Oct 2012 13:18:45 GMT

Thanks Mike

Mike conclusion after studying this interesting post is that I have to go to the optician very sooooooon.

Looking forward to your next post,

Julian.

mikeash - 2012-10-15 02:43:26

Mon, 15 Oct 2012 02:43:26 GMT

Julian Iaccopucci: Well, here's the actual function prototype:

OSStatus ExtAudioFileSetProperty (

   ExtAudioFileRef         inExtAudioFile,

   ExtAudioFilePropertyID  inPropertyID,

   UInt32                  inPropertyDataSize,

   const void              *inPropertyData

);

No AudioFileID to be found.

Julian Iaccopucci - 2012-10-14 14:16:29

Sun, 14 Oct 2012 14:16:29 GMT

I just went trough the entire program trying to understand and learn every "bite" of code. I can be wrong, but I thought that ExtAudioFileSetProperty takes as first parameter an AudioFileID data type.

Thanks Mike,

Julian.

mikeash - 2012-10-14 13:21:14

Sun, 14 Oct 2012 13:21:14 GMT

Julian Iaccopucci: I'm not sure I understand what you're asking. If I'm using ExtAudioFile for everything, why would AudioFileID or AudioFileOpenURL be involved?

Julian Iaccopucci - 2012-10-13 19:11:01

Sat, 13 Oct 2012 19:11:01 GMT

Hi Mike.

In order to set the properties with ExtAudioFileSetProperty shouldn't we need to pass AudioFileID created with AudioFileOpenURL instead of an ExtAudioFileRef created with ExtAudioFileOpenURL?

Thanks for the always pleasent read,

Julian.

jamie - 2012-10-13 16:39:14

Sat, 13 Oct 2012 16:39:14 GMT

Naturally you put this up today, after I spent the week writing <a href="https://github.com/iluvcapra/JHWaveform">MY waveform overview class</a>.

A subjects that might follow this is a metering NSView class.