voiceloop: the fastest voice agent loop in the browser is now open source

Описание и применение

voiceloop — открытая библиотека JavaScript, обеспечивающая быстрый (862 мс) и прерываемый голосовой цикл в браузере без облачных сервисов, с фильтром эха и полной локальной обработкой.

Задача

Building a low‑latency, interruptible voice agent in the browser was difficult due to competing open frameworks that either cut off the user’s voice or required cloud APIs.

Результат

voiceloop achieves 862 ms median latency with 1067 ms p95, outperforms OpenAI Realtime and Pipecat, and eliminates self‑interruption and echo issues while remaining fully local and open source.

Рабочий процесс

• 1. Install the library via npm: npm i @todoforai/voiceloop
• 2. Create a VoiceAgent instance with desired LLM endpoint, model, persona, and event handler
• 3. Call unlockAudio() on user interaction to enable microphone
• 4. Start the agent with agent.start()
• 5. Handle assistant events to render text or other UI updates

Доказательства

voiceloop with Deepgram Flux + ElevenLabs flash is the fastest configuration we measured, at 862ms median — and its p95 (1067ms) is the tightest in the table by a wide margin.

Источник

voiceloop cut itself zero times and ran echo‑coupled turns at 930ms — parity with clean.

Источник