Implement CNN on DSP

[LIPING YANG]

Hi DCASE community,

      I will try to implement CNN on DSP processor for acoustic scene classification. But, I do not find a suitable DSP chip and corresponding development environment. Are there any suggestion about the manufacturies of the chips. Thank you.

----------------------------------------------------------

Best Regards,

Liping Yang

[Jon Nordby]

Hi,
Can you use a general purpose microcontroller instead of a DSP? The ARM Cortex M4F and Cortex M7 architectures are suitable for small neural networks, and has good software support for efficient CNNs. They also have several DSP extensions and SIMD.

CMSIS-NN documentation
http://arm-software.github.io/CMSIS_5/NN/html/group__CNNExample.html

They also have a good paper
CMSIS-NN: Efficient Neural Network Kernels for Arm Cortex-M CPUs.
https://arxiv.org/abs/1801.06601

Development boards are available from all the usual microcontroller vendors. One pretty powerful example is STM32F4Discovery, NRF52Thingy from Nordic is a very low power one.

I have not found any good examples of audio event detection yet, so let me know if you do. Interested in the same!

[Jose Giraldo]

Hi Liping and Jon,

Eta compute has been working on that, despite they're using SNN it seems that is possible to perform CNN as well.

https://spectrum.ieee.org/tech-talk/semiconductors/processors/eta-compute-debuts-spiking-neural-network-chip-for-edge-ai?fbclid=IwAR0TDJkJM_d0iwp7n-IsJGpNvwk17K6k_tKv7b0MfYQr34-qnuRbtDURInk

http://etacompute.com/news/press-releases/684/

I'm also interested in these topics.

--
You received this message because you are subscribed to the Google Groups "DCASE Discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dcase-discussi...@googlegroups.com.
To post to this group, send email to dcase-di...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/dcase-discussions/CAF9BRB2_mUEF5ZHizz2Uc56C4AOS%2BQEQ%3D%2Berxf0DOeahq8yZXQ%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.