Most real time applications require at least 24 bit data.   But CD quality sound is just 16 bits wide.  The FFT algorithm is easy to parallelize, so it is perfect for a many core device.   Sound localization requires many cross correlations so it is also a perfect application for an FPGA with 8 multipliers.  Furthermore many Forth developers are older, and both need a hearing aid, and would love to be able to hack on their hearing aids.  And in the US, hearing aids are no longrr regulated, and since many countries follow US FDA policies, that means that the same is true in much of the world.

The idea is to do a hearing aid based on sound localization, and frequency shifting.  Based on time of flight and cross correlations, identify where the sounds are coming from, deduct the noise from the signal, and amplify the source that the user is looking at.  Then pass that data through a frequency shifter optimized for the users hearing loss, and programmed by the user.   With open source Forth, users can make huge optimizations in the basic algorithms.

We will use two wired omnidirectional Lavalier Microphones (Lapel Mics), connected to a dilligent stereo input and output pmod.  Data will be collected at 48Khz.  480 data points will be cross correlated 100 times a second by the 8 multipliers to localize the sound, deduct the noise sources, and amplify the source in front of the listener.    The user can clip the two lavalier mics to their shoulders, and listen to the hearing aid over earbuds.  They can cary the low poer board in their backpack.

And of course to create a high quality hearing aid, the data path should ideally be at least 24 bits wide, but that is easy to do on  an FPGA, the limitation is on memory bandwidth, not on the FPGA fabric itself.

 

 Built using the  Forest Map Wiki