use case · multi-model · M2M100 + Kokoro
Two browser models in one pipeline. M2M100 translates text from any of 100 languages into English; then Kokoro reads that English aloud in a voice you pick. Paste something foreign, understand it, and hear it — and nothing ever leaves the device.
Kokoro's on-device voices are English, so this pipeline translates into English and speaks it. To go the other way as text, use the basics demo.
Each model initialises on its own below (cached copies auto-start, no button).
M2M100 — translation
Kokoro — text-to-speech
step 1 — you
step 2 — M2M100
step 3 — Kokoro
Translation and speech synthesis are each useful alone. Chained, they're an accessibility and comprehension tool: paste a foreign sign, message, or article and hear it in plain English, with no server, no API bill, and no data leaving the tab. Add speech-to-text in front and you have a spoken translator; the same building blocks compose in whatever order your app needs.