Limitations of Speech Synthesis API

lindsay-greene commented 3 years ago

The phonetic pronunciation of the web speech API is pretty inconsistent. From my explorations, here is a summary of what I have learned:

If you only type an individual letter, or multiple letters in a row that it doesn't know how to read ("ff", "zz") it will say the name of the letter however many times it was typed.
If you add a vowel to the end of a consonant, the pronunciation is varied and often inconsistent. For example, “buh” and “muh” were correctly pronounced as expected, but “puh” and “fuh” were pronounced as “poo” and “foo”.
Fricatives are probably the hardest consonants to reproduce. The api is very resistant to drawn-out sounds. The best I could do was to either put a vowel in between and then repeat the consonant ("sis", "ziz") or just put a vowel on the end to get as close as possible to the desired sound ("fa").

lindsay-greene commented 3 years ago

Potential tool: SAMPA

SAMPA aims to create a computer-readable phonetic alphabet by matching IPA symbols with ASCII characters
Did not work in VS Code either as a string or entered plainly, but more research needs to be done on specific uses and notations to see whether this has potential

lindsay-greene commented 3 years ago

Potential tool: aria-ssml

Proposal to include speech synthesis markup language, a notation for instructing the computer on phonetic pronunciation, in HTML
This looks like the tool with the most potential if it is made reality, but since it is just a proposal as of right now, it cannot be incorporated into our code

fizzstudio / phonetics-speech-api