Comparison about julius

[adefossez]

@jonashaag thanks a lot for that! By the way I just fixed an issue in the code that was particularly bad for high sampling rate in irreductible form, which is exactly the case you are covering.
The results are much better now, see screenshot here after. When you get a chance could you update your notebook with it?

[jonashaag]

Cool! :-)

If you're interested we could also make it into a proper repo and collaboratively add new implementations from time to time. In the meantime I'll update the Gist.

[jonashaag]

Btw would you mind adding benchmark column with resampy kaiser_fast (which I have found to be good enough mostly)?

[jonashaag]

https://github.com/jonashaag/audio-resampling-in-python

Not sure why it doesn't render though...

[adefossez]

I'm only replying now about kaiser_fast. Actually when doing the benchmark I am setting the number of zero crossing in julius to the same value as kaiser_best. Comparing to kaiser_fast, and updating the number of zero crossings for julius just roughly scales both algorithm runtime by the same factor, so it has not much interest.