That's a good start. However would be nice to have some automated
tests, as speech can be so subjective.
It's possible for differences to creep in to any port. Note to self -
It would be nice to have some standardised tests so we can rubber stamp
any port. I can think of places would compare vectors and verify chunks
of code at the unit test level.
This sort of thing is actually easier for fixed point code - you can
make it bit exact end-end.
- David
On 3/12/20 2:24 pm, Matt Weeks wrote:
I just tested a few samples I captured myself on both. I did notice
some minor floating point rounding differences, but it still seems to
sound fine. I just reproduced the random number generator so the
results would be the same.
On Wed, Dec 2, 2020 at 1:39 PM David Rowe <[email protected]
<mailto:[email protected]>> wrote:
Wow that's cool Matt well done! How did you verify the port?
This is always a problem with speech based systems, even verifying
the C port between platforms can be tricky as the random number
generators are different, floating point precision etc. It is
possible to compare vectors for some of the processing stages.
- David
_______________________________________________
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2
_______________________________________________
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2
