devwithjevReal-Time Voice-Agent Backchanneling
今井智章 tested Jev for real-time sentiment analysis and context-aware backchanneling in a voice agent.
View on X time300msec以内
今井智章/シリコンバレーでファウンダーCTO@tomoaki_imai𝕏
ようやく @typesafeai のjevのアクセスが来たので、早速音声エージェントのリアルタイムセンチメント分析と文脈を認識したあいづちを試してみた。めちゃいい感じ。 - 発話内容はjevが300msec以内に分析 - 100-200msecごとに会話が途中でも読み取って内容に応じて会話を促す/説明を理解する/同意するなどあいづちを返す 通常のフローはLLMにまかせつつ、細かい会話はjevの情報でハンドリングするという使い方ができそう
Jev analyzes speech content within 300msec. Every 100–200msec, it reads the ongoing conversation and responds with backchannels that prompt the speaker, show understanding, or agree. He suggests using Jev to handle conversational details while leaving the usual flow to an LLM.
Also filed under Tools & apps
- Meta Ad Analysis Chrome Extension
The Chrome extension analyzes ads from Meta’s Ads Library using JEV (typesafe/jev-1.13) via OpenRouter, surfacing result potential, angle, hook, and niche. The user’s key stays in their browser.
- Jev Plugin for the PI Harness
The repository provides a Jev plugin for the PI Harness.
- Jev Suggestion Plugin for Hermes
The repository is a Hermes plugin named `jev-suggest`.
- OpenRouter Draft Evaluator
This local prototype evaluates and compares drafts generated with OpenRouter using Jev.