How to record audio continuously, and have access to last x seconds of data??

I would like to be able to record audio continuosly, and be able to process the last say 10 seconds of data. The recording may run for a while before I need to process the data, and so I need a solution which only saves the last 10 seconds, and doesn't save data from the whole period.

Can anyone suggest a way of implimenting this? I think this may be called a 'circular buffer'

Answer 1

[jibrahim]

Jack,

you should get to what you want with audioDeviceReader and dsp.AsyncBuffer

% Create an audio device reader and set its properties based on your

% scenario

fs = 44100;

adw = audioDeviceReader('SampleRate',fs);

% Create a circular buffer.

% Set its capacity to 10 seconds

buff = dsp.AsyncBuffer('Capacity',10*fs);

while (1)

% Read from mic

frame = adw();

% Write the frame to the buffer

write(buff,frame);

end

% when you are ready, use read(buff,frameLength) to read from the buffer

[Walter Roberson]

Cool, that makes things easier!

[Jack Latham]

Thank you for the answer! I still can't impliment this for my application, I should have explained clearer. I'm trying to run a program which classifies audio in real time, so I take say 4s of data, process it and make a prediction, I then need the next 4 seconds. So I need the section in the while loop to run simultaneously to the classification section.

I currently achieve this using audiorecorder, which records continuosly in the background, allowing me perform analysis in matlab, and when required aquire the recording, and simply used the last 4 seconds (this is obviously not a great solution!)

Again thank you both for your useful input, any suggestions on how I can impliment this solution 'in the background' would be amazing.

[jibrahim]

Hi Jack,

This example might be helpful:

https://www.mathworks.com/help/audio/examples/Speech-Command-Recognition-Using-Deep-Learning.html

In particular, take a look at the last section:

Detect Commands Using Streaming Audio from Microphone

It shows how a trained network is used to classify commands in real time, as a singal is streaming in from a microphone. It uses audioDeviceReader and implements a cicular buffer using a variable waveBuffer (though you could use dsp.Asyncbuffer as well).

[Walter Roberson]

audiorecorder() is not the right tool for streaming audio, as it can only record for a pre-determined amount of time and accessing the data is while recording is running is a bit tricky sometimes.

Answer 2

[Walter Roberson]

Audio Systems Toolbox, use the device reader, and as you read the data, store it in a buffer large enough for the number of seconds you need plus the latency of responding to the user clicking to say "There! That data!"

Often the most efficient way to store data in such circumstances is with a "circular buffer" https://www.mathworks.com/matlabcentral/answers/65213-how-to-develop-a-fifo-circular-buffer-for-analyzing-real-time-data

[Jack Latham]

Thanks, yeah that's exactly what I'm trying to do, it is implimenting a cicurcular buffer (for example with audioDeviceReader or audiorecorder) which I'm stuck with.