wkfedor/deepseek-harness-voice-input
Voice Input adds voice typing, microphone dictation and local speech-to-text to the DeepSeek Harness Web UI: a microphone button in the prompt composer records speech in the browser and transcribes it locally with multilingual Whisper via Transformers.js and ONNX Runtime, across Firefox, Chrome, Edge and Safari with no extension, Python or cloud speech API.
Voice Input and Speech-to-Text adds voice typing, microphone dictation and local speech-to-text to the DeepSeek Harness Web UI. A single microphone button integrated into the prompt composer records mono PCM audio, produces a 16 kHz WAV payload, sends it only to the local Harness server, and transcribes it in a separate worker with multilingual Whisper (Transformers.js + ONNX Runtime), so no recording is sent to DeepSeek, Google, Microsoft, OpenAI or another cloud provider. It works across Firefox, Chrome, Chromium, Edge and Safari without a browser extension, Python, pip, ffmpeg, administrator access or an external speech API. The first microphone click opens an installer card that downloads the model (percentage, bytes, speed, cancel, retry), setup is idempotent and persists across page reloads, and the plugin only prepares text - it never auto-submits, so you can review or edit the transcription before sending.
Install
dsh plugin --profile web add github:wkfedor/deepseek-harness-voice-inputCopied verbatim from the README's install block; the README then instructs adding an insert entry for id: voice-input / name: deepseek-harness-voice-input to $DSH_HOME/profiles/web/cordis.patch.yml (merging into the existing top-level array, not overwriting).
Compatibility
Records speech in the browser (getUserMedia + Web Audio API, not the browser SpeechRecognition service) and transcribes locally with multilingual Whisper via @huggingface/transformers and ONNX Runtime; the audio payload goes only to the local Harness server and is deleted after transcription. Uses onnx-community/whisper-tiny q4 multilingual weights (~100 MB), cached under $DSH_HOME/voice-input/models. Intended for current Firefox, Chromium, Edge and Safari builds. English README with a Russian section.
Details
- Repo: wkfedor/deepseek-harness-voice-input
- Category: Vision & Multimodal
- Stars: 1
- Version: GitHub repo wkfedor/deepseek-harness-voice-input; installed from GitHub (no npm package is documented). The README labels the package a Cordis plugin rather than a profile bundle.
Alternatives
54xkeee/dsh-vision · 54xkeee/dsh-youreyes · Cheng-cheng9669/dsh-deepseek-vision