DeepSeek Harness Local Voice Wake-Up and Speech Recognition Plugin
The DeepSeek Harness (DSH) plugin `dsh-koein` provides a local voice input solution that requires no cloud interaction. The plugin uses a local keyword spotter (KWS) to wake the agent and local offline automatic speech recognition (ASR) to convert speech into text. It is characterized by zero dependencies, zero API keys, and zero manual input, making it especially suitable for privacy-focused environments or local deployments where cloud APIs are unavailable. Installation and usage should refer to the source code or documentation, and attention should also be paid to permissions and licensing issues.
Read Moredsh-funasr-voice: Local speech recognition in DSH input box
`dsh-funasr-voice` is a local offline voice input plugin for DSH Web, designed to enable speech-to-text without external interfaces via local ASR. The plugin utilizes an architecture where microphone capture is performed on the browser side and FunASR + SenseVoiceSmall runs on the Host, supporting approximately 10x real-time speed recognition on local CPUs. It features automatic punctuation, live captioning, real-time display of recording duration, and automatic insertion into drafts. Installation requires adding it via the DSH Plugin Manager, followed by one-click dependency installation (including venv, funasr, torch, and model downloads) on the settings page, or manual configuration of the Python interpreter and model paths. Configuration supports either a GUI or YAML files, allowing customization of behaviors such as language, port, and auto-send. This plugin...
Read More