Shizuku AI Runs an On-Device iPhone Voice Demo Without the Cloud
Shizuku AI Technology Department released a prototype demo on November 5 in which the AI VTuber Shizuku holds voice conversations using only an iPhone. Speech recognition, replies, and voice synthesis all run on the device.
1 source 4h ago
Reporting from 1 source: ASCII.jp.
Why it matters
The point of the experiment is the tradeoff it makes visible: dropping the cloud gets Shizuku onto a phone, but the model small enough to run there talks worse than the streamed version.
Key Facts
- Shizuku AI Technology Department released the on-device voice demo on November 5.
- The prototype runs speech recognition, language model replies, and voice synthesis on an iPhone without the cloud.
- The demo uses the iPhone's GPU and Neural Engine to speed up responses.
- The department says the on-phone language model has lower conversational ability than the usual Shizuku.
- Shizuku-Duplex, a Japanese voice dialogue model, was announced on November 8 and processes audio every 80 milliseconds.
The demo uses a lightweight on-phone language model, and the department says its conversational ability is lower than the usual Shizuku. On November 8 the department announced Shizuku-Duplex, a Japanese voice dialogue model that processes audio every 80 milliseconds.
The demo swaps typed comments for speech. A user talks to the phone, recognition turns the audio into text, a language model on the device drafts a reply, and voice synthesis reads it back in Shizuku's voice. The iPhone's GPU and Neural Engine, the chip's AI accelerator, are used to shorten response times.
Shizuku AI Technology Department framed the prototype as a question about scale. As LLMs grow larger, staff asked how naturally a conversation could be held on a smartphone alone, and accepted a smaller brain for the test. The department described the result as a clumsier Shizuku than usual.
The Duplex model announced on November 8 targets a different problem. Full-duplex processing lets it listen while it speaks, and complex searches are pushed to a separate model behind the scenes.
Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google