Human voice discrimination system : board?

[ank]

Hi all, 

I am planning on implementing a human voice discrimination system on a board. Just for fun. I have already implemented such algorithms before, and their performance is fine. 

Before I figure out how to implement something like that on hardware, I want to purchase a decent board <$300 to accomplish the project. I am determined to figure it out. I just want someone with experience to tell me if ZYBO Z7-10 be enough, or if I should purchase the ZYBO Z7-20. Link to the product here. If neither of these may be enough please do suggest anything better. 

Thank you for your inputs. 

Regards,

Ankur  

[D@n]

@ank,

Hard to say.  Depends upon your algorithm.  Are you intending to run the algorithm in the FPGA portion of the board?  Does it use fixed point or floating point?  How many multiplies will you need?  etc., etc.

Dan

[zygot]

2 hours ago, D@n said:

Hard to say. 

I wasn't brave enough to be the first to reply to this one.

I do like the way that @D@n has started things though. The ZYNQ 7020 has a PL that is roughly equivalent in resources to an Artix 75 device. The 7010 would have fewer PL resources. If you aren't going implement most of your data crunching in logic then I'd suggest skipping the FPGA device altogether and just using a high end micro or even a recent Raspberry Pi 4. I'd definitely not waste time trying to figure this out on an FPGA based platform without first implementing it in software and being convinced that it's worth the extra effort. The proper place to do this for a first cut would be on a PC. You can always work on acceleration of specific parts where needed later.  If all you want to do is try and implement a few algorithms in logic then start with memory resource requirements. I'd think that just about any recent board would do. My preference for such a case would be to use a board like the Terasic C5P. You can stick it into your PC and use PCIe to transfer data between PC applications and logic applications with little consideration of data transfer rates.

Just curious. What's the impetus and motivation for such a project? You mention that you've implemented such a system before and had fine results. What do fine results mean in real world testing? I'm not judging, just find the whole idea a bit odd and curious. These days, I wouldn't even care to offer a public definition for the term 'female voice'.