Hi everyone,
As some of you on the more recent Mac OS versions might have noticed, Tcl/Tk+snack can be a legacy nightmare, at least it was for me. Which got me thinking, the pitch extraction script (getf0.tcl) in the HTS demo uses the RAPT algorithm, which is also present in SPTK. Since SPTK is already required by the project, I tried to see if it was possible to cut out Tcl altogether.
After modifying the configure and makefiles a bit, I was able to obtain good results. I've inspected several files, and I'm sharing figures for cmu_us_arctic_slt_a0001.raw
Here's the overall comparison:
http://i.imgur.com/J7uZQrK.png
And here's the close-up of the part with the most discrepancy:
I think the discrepancy is very small, though I don't know why it happens, slight implementation difference between snack and SPTK maybe? I'd really love to hear if someone with more knowledge on getf0.tcl would weigh in on this.
What do you all think? I haven't done a comprehensive analysis yet, but if the discrepancies are insignificant (or maybe SPTK is actually better?), getting rid of Tcl/Tk and snack would not only cut out complexity for beginners, but more importantly would enable deployment on more platforms without user tinkering.
Best,
