okay thanks! The default number of formants (n_formants) is not 4? fasttrackiverse.github.io/fasttrackpy/...

I'm just going by what worked in Praat, which was to change the formant settings to 4 formants with a max of 5 kHz. Should changing max_min_formant to 5000 not have a similar effect?

Looks like o was worst among our comparisons, but PRICE was also bad:

e.g. extremely fronted MOUTH tokens (female speaker) where the correct F1,F2 are like 713 and 2215 (via DARLA, checked in Praat), but new-fave is giving 675 and 876 using the defaults. Phonologically front vowels seem to track OK.

We're going to try upping the min_max_formant parameter?

cool! we're gonna try tweaking the config file, and will keep this in our back pocket...

new-fave I believe uses FastTrack's methods. the issue we have is that archival (crappy) audio seems to get decent results from DARLA (with zero filtering!) but be error-prone in new-fave, while high-quality new audio from speakers with the same dialectal back-vowel hijinks tracks fine in new-fave.

I'll be looking for your tips and tricks!

if you're thinking matched guise, I'll give you three guesses as to the initial L. in my name 😉

opaque 😉

there's a line of research out of UIUC on this, and they frame it as a marginal contrast. there are several papers. here's one: www.frontiersin.org/journals/com...

A house in my neighborhood had a banner reading TR*MP 2024: MAKE LIBERALS CRY AGAIN... yeah a few weeks ago it finally disappeared