High latency for audio use on a Raspberry Pi using Simulink
10 views (last 30 days)
For a student project we decided to build test hardware for musical instrument effects. As base we decided to go for a Raspberry Pi with some peripheral equipment installed. We've programed the hardware just using Simulink with their tools and libraries. Everything works fine but we do have a huge latency with our audio signal.
Even when we try to put the signal through without effects, the latency is half a second long.
Do you have any ideas why?
Does the Raspberry Pi compute on its own or is all data just sent via ethernet?
How do I reduce buffer sizes?
Thanks for your help!
Tom Haelvoet on 8 Oct 2016
Edited: Tom Haelvoet on 8 Oct 2016
The latency can be reduced from within Simulink after performing the following steps:
1. Drag & drop an ALSA Audio Playback block into your model.
2. Jump to the Library (CTRL+L) and unlock it.
3. Edit the mask (CTRL+M) of the ALSA Audio Playback block.
4. Navigate to Parameters & Dialog, Select Que duration & check the Visible flag.
5. Lock & save the library.
The Queue duration should now be visible and ready to be modified.
FYI: Tested with a Focusrite 2i2 (Device name = 'plughw:1.0') and a Queue duration of 50ms
Murat Belge on 10 Dec 2014
The ALSA Audio Playback block uses and audio buffer size of 0.5 seconds of data. You can not change this parameter unless you modify the source file (MW_alsa_audio.c). This parameter determines the latency.
It is very hard to achieve real-time, low-latency audio in Linux. If you reduce the audio buffer size, you might observe stutter / interruptions in the audio.
Dong-uk Hyon on 1 Mar 2018
On top of Tom's answer above, I am also carrying out audio experiments using Raspberry Pi and have found lowering the ALSA frame sample size down will significantly reduce the latency.
I found these settings by double clicking on the ALSA audio capture block. The default frame size for me was 4410 (N), by reducing this value you will find a better response of your system. I managed to go down to 10, and the delay is just about noticeable by audio.
Info found here: https://uk.mathworks.com/help/dsp/ug/sample-and-frame-based-concepts.html#bso3wxr-1
Kind regards, Don