dsh-voice

DeepSeek Harness is a powerful text-based AI assistant, but its API only supports text interactions, lacking the ability to directly handle audio input or generate speech output, which limits its use in voice-centric scenarios. The dsh-voice plugin addresses this by adding comprehensive voice capabilities to DeepSeek Harness, enabling users to talk to the model via a microphone and have it read replies aloud, thereby extending interaction methods without altering the model itself.

The core abilities of dsh-voice include voice input (STT) and voice output (TTS) on the web interface. Users can click a microphone button to record speech, which is transcribed to text using the browser's Web Speech API and a configurable Whisper-compatible backend; each reply has a speaker button that reads it aloud via speechSynthesis. Additionally, the plugin provides voice_transcribe and voice_speak agent tools for transcribing audio files and synthesizing text to speech. Notably, all these features require no API keys and can be paired with free backends like local whisper.cpp for a fully offline, keyless setup.

This plugin is suitable for a wide range of users. For everyday users, it offers a more natural and convenient voice interaction method, especially useful when typing is impractical. For developers and AI enthusiasts, it demonstrates how to extend AI assistant capabilities at the client edge, making it ideal for local deployment or experimentation. In education or assistive technology fields, dsh-voice enhances accessibility, helping visually impaired users or language learners interact with AI more easily.