Overview
dsh-voice
ChatGPT-style voice input for the DeepSeek Harness Web GUI: mic button, recording pill with live waveform, transcription via transcribe.cpp (Handy model catalog), whisper.cpp, the Codex/ChatGPT subscription, or an OpenAI-compatible API
README / EN
Package documentation
Registry summary
ChatGPT-style voice input for the DeepSeek Harness Web GUI: mic button, recording pill with live waveform, transcription via transcribe.cpp (Handy model catalog), whisper.cpp, the Codex/ChatGPT subscription, or an OpenAI-compatible API
dsh.pub verifies the pinned bundle contract, runtime facts, and distribution semantics. The complete README remains in the source repository.
Read the full README on GitHubLIMITATIONS
Known limitations
- Recording uses the browser `MediaRecorder` (HTTPS or localhost required). - Transcription is batch (after the recording ends); streaming partials are planned. - Upstream whisper.cpp does not support Handy's Qwen3-ASR GGUF (needs the transcribe-cpp engine); standard ggml models (tiny…large-v3) work.