dsh-realtime-voice: Full-Duplex Voice Agent Plugin
`dsh-realtime-voice` is a full-duplex voice agent plugin designed for DeepSeek Harness to simplify voice agent integration. It provides a provider-agnostic abstraction layer that decouples developers from underlying provider protocols, shielding complex details such as OpenAI WebRTC and Doubao WebSocket. It supports real-time dialogue, barge-in logic, and browser audio management. Developers can start a session by calling a unified API, significantly reducing the development barrier.
Read MoreDeepSeek Harness Real-Time Voice Plugin (dsh-realtime-voice)
The DeepSeek Harness plugin `dsh-realtime-voice` enables real-time voice interaction between browser audio and LLMs. The plugin supports dual routing for Alibaba Cloud Bailian and OpenAI, uses a same-origin WebSocket proxy to provide automatic speech recognition (ASR) and text-to-speech (TTS), and features streaming audio playback with a candidate interruption review mechanism. After installation, users can use the plugin by configuring the corresponding API key in Harness. Considerations include microphone permission limitations in the desktop Electron version, strict same-origin security policies, and a voiceprint feature intended only for suppressing background speech.
Read More