Last time out, I discussed a better, more systematic path to finding homophones: convert words and phrases to their pronunciations and get them into a spreadsheet.
But to do that with a big resource—like, say, my personal list of film titles—I’d need to find a transcription service that (1) could handle large amounts of data and (2) didn’t stall on stuff like “Carrey” and “tae bo” that’s not in the CMU pronouncing dictionary.
I don’t think that’s unreasonable. I’m not asking for 100% accuracy when reading “Mxyzptlk” or a normal phrase with the spaces removed like IINSIST (I insist). But the service should at least try. To pick out fresh homophones from a big collection of data, it has to not fear the unexpected, and it has to scale.
Googling “convert text to IPA” gets you a variety of online services. To start with, I tested each on the sentence, “Jim Carrey does tae bo at the gym with Carrie Fisher.”
Top search result ToPhonetics failed. Phonetizer failed. A “Convert to IPA” program on GitHub, the third Google result, was nonfunctional. EasyPronunciation.com failed because the text was longer than 25 characters(!). The IPA Translator app for Google Docs and Google Sheets looked promising but did not work at all.
GlotDojo succeeded. EnglishTools.org succeeded. InternationalPhonetcialAlphabet.com succeeded. UnaLengua succeeded…
And none of them scaled. UnaLengua was kind enough to warn me that it’d only take 2,000 characters; the others revealed their limits soon enough.
The only service that scaled close to the level I needed was IPA Now from PhoneticSoft. It advertised it could do transcriptions with no limit. My testing found that claim exaggerated, but it did IPA-ize thousands of words at a time, enough that—with a few hours’ patience—I could get my movie list done. I published the results of that work as “Homophone Double Feature” on this channel.
IPA Now was not a free service, but it did allow me a free seven-day trial. I’ve thought about subscribing for the sake of further research, but the service does have one defect aside from its aforementioned limit: it transcribes with a British accent.
Other convert-to-IPA systems use standard British pronunciation or toggle between standard British and standard American. Even “standard American pronunciation” is a wild generalization. Pronunciations vary between American regions, and they’re changing all the time. Still, as a ’merican myself, I tend to prefer a service that can distinguish between “Barbie” and “Bobby.”
On and on it goes, an endless parade of not-quite-right. I thought a downloadable program might be the answer, but some are designed for mobile phones or phone emulators, while others specify a size limit or a dictionary limit…or fail to mention any such limits but have them anyway. (The CMU pronouncing dictionary is not the answer to everything!)
My issues here echo one of my concerns about AI: we seem to be rewarding programs that are not so much truly “intelligent” as good enough to look smart. Maybe these programs are enough if you’re using them for less eccentric purposes than mine. But the reality falls short of the promise.
IPA Now is the best of this bad batch, by far. So I’ll probably subscribe to it, despite its limitations…unless you, dear readers, can suggest something better. I know some readers of this feature are plugged into the word-study scene; you may know things that I do not. I’m always saying “Feel free to tell me what I’ve missed in the comments.” But today…feel extra free!