Closed tdjastrzebski closed 11 months ago
@tdjastrzebski There is currently no such algorithm. But the CMSIS-DSP kernels can be used to build some.
To compute the ZCR itself you could multiply the signal with a delayed version by 1 sample. You may use CMSIS-DSP to remove any DC bias from the signal.
As an example of use of the CMSIS-DSP Python wrapper I had created this notebook:
It still use NumPy for the zero crossing counting. But other parts (filtering, classifying etc ...) are done with CMSIS-DSP. The zero crossing could partially be done with CMSIS-DSP too.
The above example is not a good example of keyword spotting and don't really work. It is just an example to demonstrate a more complex use case and you may find some ideas to reuse in it.
@christophe0606 Thank you for sharing this solution!
Does CMSIS-DSP implement any algorithm which can be used for zero-crossing detection (ZDC)? I understand that it is a broad question and zero-crossing detection is a science in itself, but I will appreciate any suggestions on whether or not CMSIS-DSP is the right tool for such task.