Chrome AI DEV

[Sock Chen]

Hi Chrome AI Team, 

I am also an early developer looking into google AI powered chrome extensions. Here is a walkthrough of my envisioned chrome ai extension, would be good to advise on the design as the audio apis may not be ready for usage. 

🎙️ Polyglot Pronounce – Your AI-Powered Language Coach in the Browser

Polyglot Pronounce is an AI-powered Chrome extension that helps you speak more clearly and write more confidently — across Malay, English, Chinese, and Tamil.

🔊 Speak. Record. Improve.

Record your voice directly in the browser and get instant, intelligent feedback on your pronunciation:

• Learn standard Malay pronunciation (localised Asian accents)

• Practice native-level English, Mandarin, or Tamil

• Get personalized tips on clarity, fluency, and accent improvement

[Thomas Steiner]

Hi Sock,

I doubt that the current APIs would deal well with this. Let's say you mispronounce the word pear (the fruit 🍐, pronounced like bear) as "peer". The API would very likely just transcribe it as 'peer' and not flag the pronunciation error. 

Cheers,

Tom

--
You received this message because you are subscribed to the Google Groups "Chrome Built-in AI Early Preview Program Discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to chrome-ai-dev-previe...@chromium.org.
To view this discussion visit https://groups.google.com/a/chromium.org/d/msgid/chrome-ai-dev-preview-discuss/15aba9ac-56b9-4ff2-9b09-32f4ec6492a0n%40chromium.org.

--

Thomas Steiner, PhD—Developer Relations Engineer (blog.tomayac.com, toot.cafe/@tomayac)

Google Spain, S.L.U.

Torre Picasso, Pl. Pablo Ruiz Picasso, 1, Tetuán, 28020 Madrid, Spain

CIF: B63272603

Inscrita en el Registro Mercantil de Madrid, sección 8, Hoja M­-435397 Tomo 24227 Folio 25

----- BEGIN PGP SIGNATURE -----

Version: GnuPG v2.4.8 (GNU/Linux)

iFy0uwAntT0bE3xtRa5AfeCheCkthAtTh3reSabiGbl0ck

0fjumBl3DCharaCTersAttH3b0ttom.xKcd.cOm/1181.

----- END PGP SIGNATURE -----

[Nikita Malyschkin]

Hey 

That sounds like a great idea!
Thomas has a point, mispronouncing so it sounds like a different word might be a major issue.
Maybe you can pivot a bit and provide the AI the text you're reading as well as the sound file.
I'm not sure if this works but you could try this out easily on a bigger model like gemini2.5. If it doesn't work there, it probably won't work on a nano model.

Super interested if this works 😄