@EricFos I could see that happening, yes.. it would be interesting to see work that *directly* tests the relationship between G2P/pronunciation lexicon and forced alignment 🤔 And great, I'll let you know after I write up my findings!
@EricFos Thanks for all your thoughts, this sounds quite relevant! Would you be able to point me to 1 or 2 example papers about what you're describing here?
@EricFos I'm looking to measure against the actual speech, and you're right, I expect the sources of variability to be many! I'm not trying to directly evaluate G2P, but indirectly (after read speech has gone thru G2P, forced alignment, and formant extraction).
PSA to anyone attending #NAACL2022: go get yourself a matcha parfait 🍵🍨at Nana's Green Tea some time this week :) It's just a couple blocks from the conf venue!
https://t.co/8FxPJwmEEo
“NLP is better for its partnership with linguistics, because linguistics grounds NLP as an application area where there is deep scholarship around the shape of the problems we solve and the social contexts our technology enters.” - @emilymbender at #NAACL2022
New paper out in Language @LanguageJofLSA with Colin Wilson! We investigate constraints on phonetic realization––the projection of phonological representations into phonetic space 🧵
https://t.co/EbdmW5yFq4
Tomorrow at #LabPhon18, I’ll also be talking about my adventures in cross-linguistic phonetic analysis, discussing uniformity in talker- and language-specific phonetic realization, and our work on developing cross-linguistic phonetic corpora (check it out via Whova)
Many of the VoxCommunis acoustic models are available via the Montreal Forced Aligner (thank you @wavable!) and OSF (https://t.co/xQ6UeLTpVR). Improving these resources for cross-linguistic phonetics will be a true community effort, so please don’t hesitate to get in touch!
#LREC2022 paper this afternoon! @emilypahn and I will be presenting VoxCommunis, a cross-linguistic phonetic corpus with phone alignments, phonetic measures, and acoustic models based on the multilingual Common Voice corpus
Paper: https://t.co/iyyRXydYxZ