Cirrus Logic - RPI 2 FFT on streaming audio

Question

Hi everyone,

My current project is to run constant FFT on an audio stream and output the data to a 3D LED Array. Google 3D LED cubes if you don't know what they are. My cube is more of a rectangular prism, 24x8x8=1536 LED's. I am new to the world of PI and have found the support for the cirrus logic card to be vague, if it wasn't for ppl like ragnar.jensen I'd probably still be stuck (thanks for all the information you've shared).

Anyway, currently I am recording 1 second of audio after a 1 second delay. So it takes 2 seconds to get the data / audio.

I then perform the FFT on 16384 samples. But this is only 1/3 of the 44000 raw audio samples recorded in the 1 second interval. I then group the frequencies into bands and then scale the amplitudes of those bands into a single byte which represents the bands power. The result of each FFT is 24 bytes, one byte for each band. All this takes 0.1 second.@@@@@

It then takes 0.004 seconds to send that data to my PIC, which I had to bit bang through the GPIO's since I couldn't get I2C or UART working.

The LED array takes the data and displays the 24x8 bits on the front row, when it receives the next group of data the front row shifts back one row. The LED array displays 3 dimensions, time (each group of data), frequencies and amplitude.

Question:

In the arecord command, I would like to erase the delay or change it to 0 seconds and change the minimum record length so I can stream the data at a faster rate to the LED array. Does anybody know how????

Answering this would result in a real time graphics equalizer on live audio and a more fluent transition between groups of data (more like a wave travelling away from you).

If anybody has an idea on how to achieve this please reply, or share your thoughts.

shabaz · Answer

Hi Daniel, Looks like a fun project : ) I'm not knowledgeable in Python but you might be pushing the limits of it to be honest. It could be better to code this in C, because there is a C API to the ALSA framework, which should allow you to read bytes from the audio device (hw:0 or whatever). Maybe there is an API for Python too, I just don't know it. I've not used the C API, but it is described on the ALSA site. Then you should be able to read exactly what you need (e.g. a fraction of a second instead of having to read a whole seconds-worth), and using C would also allow you to maybe do your FFT faster (just a guess). If you really want to speed the calculation up, then there is a feature called 'NEON' on some of the ARM processors (not on the older RPI models, but is available on the RPI 2 and BeagleBone Black) but it takes some learning to use. Fortunately you don't need to learn it, there are libraries out there that are already coded to take advantage of NEON. Also, you may wish to migrate to the BeagleBone Black since there is a Music Tech design challenge for it out there, if you (say) want to convert your device into a tool for creating and visualising sounds : ) Just an idea! Less than two days left to apply for it though.

balearicdynamics · Answer

Hi Daniel, I have already experienced a similar problem using the CL for data processing in the Meditech project. One of the difficulties is to download the RT Linux for PI. I have a copy and if you need I can send you it by private email just because I have searched it again recently and I have not found About the issues on the RT Kernel specific for the CL and some details I have published some articles in past that I will link below. More details I can find in my local archives. Meditech: Microphonic stethoscope Cirrus Logic Audio Card: audio monitor while recording Meditech: Text-To-Speech user support Here and there in these posts there are some useful information, links and also good comments from the other valuable guys. For more detailed information let me know because to have the stuff working I had to do a lot of research and tests and I have saved almost all the documentation. Hope this can help. Enrico