How to get sound frequency data in real time, with android.

The class I use for recording audio in real time is called android.media.AudioRecord.
http://developer.android.com/reference/ ... ecord.html

It can record in real time, the only problem was getting the frequency data from the byte stream it returns. The AudioRecord class PCM encodes the audio data, so you need to decode the PCM format to get the frequency. There are several methods for decoding PCM data, the easiest for me was called Counting Zero Crossings. I've implemented a AudioRecord threaded class below that will update a variable in real time with the frequency currently "heard" in the microphone:



One thing that might give you a headache, is that the emulator only works with certain settings when instantiating your AudioRecord. There are options you can use for bit-rate, channel, and format. As of now only the ones I used in the example will work on the emulator. Other options should work on phones though.

Hi,

thanks for those lines of code: i guess they'll prove handy to me, although at the moment, I'm starting getting that famous headache: for the moment, I cannot get the audio record state_initialised as true on my phone (htc desire)...
I guess I'll have to fiddle with the settings to get it on...

pfff : it took me a bit of fiddling to finally understand this: I am always forgetting the bases. Stupid me.



A little of that in my manifest solved my initialisation issue...

Thanks for your class, it'll make a great start for my stuffs

Oh darn, should have mention that before. Glad you found it. Btw, I'm not to sure if the formula I used to get the actuall frequency is correct, it's kind of sporatic. The frequency should be 1 divided by the time for one cycle. So I used 8000 / buffersize for the 1 second. Since in the example we are recording at 8khz. And the (numCrossings/2) should give you how many cycles per buffersize samples. Let me know if it works for you. Also to find the volume I think you multiply the numCrossings times your buffersize. But you have to have a reference to compare it too.

Nevermind it was a stupid question

Hi

This is what i was looking for for so long. One question this mite be a stupid question...

Why do you lookp every 4 to find the zero crossing point.

Hi

Here is another one

Whats the difference between using Counting Zero Crossings and FFT. I did a quick google search and if i want to do and understand fft i must go to learn some advanced maths because dam its way over my head.


thanks

fbarnard wrote:Hi

Here is another one

Whats the difference between using Counting Zero Crossings and FFT. I did a quick google search and if i want to do and understand fft i must go to learn some advanced maths because dam its way over my head.


thanks

Which is why I used zero crossing's method. The main difference is that the FFT will give a more precise and accurate frequency, especially if theres mixed tones. The zero crossing is less precise and gives the overall tone.

fbarnard wrote:

Nevermind it was a stupid question

Hi

This is what i was looking for for so long. One question this mite be a stupid question...

Why do you lookp every 4 to find the zero crossing point.

Looping every 4 is a technique called loop unrolling. It is a bit faster than looping every 1. It still works the same, as I count each sample. It is faster because you spend less time doing the loops (incrementing and checking for end of loop). The second loop after the first one, is encase the buffersize is not evenly divisible by 4, to process the remaining samples.

thanks...

Hello,
Im currently working on a project that requires me to check if a certain frequency tone (Created using the ToneGenerator Class) has been heard on an second android devices microphone. I have currently been researching FFT but I am extremly new to DSP. I have a general understanding of fft but thats all.

Would I be able to use this code to find a certain frequency (I understand there can be inaccuracies due to tones mixing and harmonics)

Thanks very much for your time and apoligies for the poor understanding of DSP.

Hello,
Im currently working on a project that requires me to check if a certain frequency tone (Created using the ToneGenerator Class) has been heard on an second android devices microphone. I have currently been researching FFT but I am extremly new to DSP. I have a general understanding of fft but thats all.

Would I be able to use this code to find a certain frequency (I understand there can be inaccuracies due to tones mixing and harmonics)

Thanks very much for your time and apoligies for the poor understanding of DSP.

This zero-crossings method is not as accurate as FFT, but is alot easier. When I researched how to decode PCM endoding, lots of sites mentioned FFT as a more accurate method, but it was hard to understand how to do it. The zero-crossings method will give you the overall approximate tone in most cases, FFT will be more accurate with mixed tones and harmonics.

thanks xdebugx. Do you have any links that explains how zero crossing works in a simple manner. I cant seem to find any basic guides about it online. Your codes very nice but when it comes to your ForLoops for the numcrossing value i dont no whats being done and would linke to understand whats happening here. If you do have a link that may explain it it would be much appreciated.

Thanks

It was hard to find examples I could understand , but this is one of the best: http://www.dsprelated.com/groups/matlab/show/2667.php

Basically your just looping through the data and counting how many times the numeric values cross zero. Or keeping track of when the sign changes. The way I did it was to check for when the data went from greater than zero to less than zero, or less than zero to greater than zero. I've simplified my code, without the for loop optimization below:

Thanks xdebugx. Last question. I understand when your trying to reconstruct the frequency using half the value of numCrossing multiplied by the number of samples your buffer is per second. But how does 8000 fit into this.

For example if I was doing a zero crossings method on a tone that is only one second long and is passed directly to the for loop which calculates the numCrossing value as an array, to reconstruct the frequency would I have


Thanks very much.

KeithB wrote:Thanks xdebugx. Last question. I understand when your trying to reconstruct the frequency using half the value of numCrossing multiplied by the number of samples your buffer is per second. But how does 8000 fit into this.

For example if I was doing a zero crossings method on a tone that is only one second long and is passed directly to the for loop which calculates the numCrossing value as an array, to reconstruct the frequency would I have

Syntax: [ Download ] [ Hide ]

Syntax: [ Download ] [ Show ]

Using java Syntax Highlighting

1. frequency = (numSamples) * (numCrossing/2) ;

Parsed in 0.035 seconds, using GeSHi 1.0.8.4

Thanks very much.

Silly me, I figured this out. Looking at code too long really does not make things easier.