WizisCool/dsh-ears
Voice input plugin for DeepSeek Harness (dsh): a microphone button in the composer turns speech into a draft transcript, with a choice of speech-recognition backends, optional polish through dsh own LLM routes, and a native settings page.
Install
via npm
Resolves a published tarball rather than cloning the repository, and installs without any extra setup. Swap `web` for your profile name if you run another one.
via GitHub · npm package
Installing from GitHub runs the project's build script, which pnpm blocks until you allowlist it — run the command once and pnpm prints the exact key to add under `allowBuilds` in ~/.dsh/profiles/web/pnpm-workspace.yaml.
What happened when we ran it
Installed cleanly when we ran it
Every command here is run in a throwaway container against a clean profile, and the result is whatever the harness recorded — not a guess from the source. Last run 24d ago.
Show it in your README
For maintainers: the badge serves this listing's latest sandbox verdict, so a re-run updates it on its own — and it links readers to the full result here.
Due diligence
What it is — dsh-ears provides voice transcription and optional polishing for DeepSeek Harness, supporting multiple backends and outputting editable drafts.
Who it is for — It suits users performing text tasks in the browser environment. Users primarily using keyboard input may not need it.
Watch out — A refresh of the Web UI is required after installation to display the microphone icon. Sandbox testing confirmed functionality in a new profile. No obvious issues were found.
The verdict — I will install it because it allows retaining the original transcription and manual editing of drafts.
Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.
dsh-ears is a DeepSeek Harness plugin that adds voice input to the web interface and turns speech into an editable draft. After installation and a Web UI refresh, a microphone icon appears beside the input box. The default shortcut is Ctrl+Shift+Space. Recording is transcribed by browser Web Speech, the local whisper CLI, Groq, Aliyun Bailian, or a specified OpenAI-compatible /audio/transcriptions endpoint. The draft is inserted into the composer for review and must be sent manually.
The host handles provider requests and local Whisper execution. In the plugin settings, dsh-ears stores the selected transcription provider and its model or endpoint details, while the optional polishing step uses a model already configured under dsh → 设置 → 模型. It stores the polishing prompt and reuses the existing LLM key rather than adding a separate key. Leaving the prompt empty uses the built-in default; failed or cancelled polishing keeps the original transcription.
This is for DSH users who want to dictate text without giving the agent automatic permission to speak or submit messages. It is a poor fit where audio must remain entirely local but local openai-whisper is unavailable, or where a Chromium-based browser is not available for Web Speech. Remote backends send recordings to their providers, and local Whisper transcribes only after recording stops. The plugin requires DeepSeek Harness 0.1.0-rc.6 or rc.7 and Node.js ^22.19.0 || >=24.0.0.
After the Web UI is refreshed, the plugin adds a microphone icon beside the input box. Use Ctrl+Shift+Space by default to start voice input. The flow is:
microphone → transcription → optional polishing → editable draft → manual send
The draft is not sent automatically. Polishing can remove speech fillers, correct ASR errors, and handle spoken corrections such as “not A, but B” and numbered points. If polishing fails or is cancelled, the original transcription remains.
| Backend | Operation | Required setup |
|---|---|---|
Web Speech |
Browser performs real-time recognition | Chromium-based browser; audio may be processed by the browser vendor |
本地 Whisper |
Host runs the local whisper CLI after recording stops |
Pre-installed openai-whisper; download a model in the plugin settings |
Groq |
Host sends the recording to the Groq Whisper API | Groq API key |
阿里云百炼 |
Host uses DashScope synchronous transcription | HTTPS origin, API key, model name; 300-second per-request limit |
自定义 OpenAI 兼容 |
Sends a POST request to the configured endpoint |
Endpoint URL, API key and model name; endpoint must provide /audio/transcriptions |
Whisper model weights are not bundled with the plugin. Provider quotas and free allowances are subject to the providers’ current documentation.
The polishing model is selected from models already connected in dsh → 设置 → 模型. The plugin stores the provider, model name and prompt, and reuses the LLM key from the existing DSH configuration. An empty prompt selects the built-in default prompt, which can be viewed in the settings page.
The documented runtime requirements are DeepSeek Harness 0.1.0-rc.6 or rc.7, and Node.js ^22.19.0 || >=24.0.0. Remove the installed plugin with dsh plugin --profile web remove dsh-ears; refresh the Web UI afterwards. For local development, the repository documents pnpm install, pnpm check, pnpm test, pnpm build, pnpm dev:config, pnpm dev:web, and pnpm dev:watch.
Written from the project's own documentation and kept in sync with it. Where the two disagree, the source is authoritative — read the README on GitHub
Same category
PolinniZhong
In-chat read-aloud for DeepSeek Harness: tap to read, pause and resume AI replies with natural Doubao TTS voices (BYOK), reading only the final answer with code, tables and diagrams filtered; local engine, plugin keeps no API key.
Installed cleanly when we ran it
PensiveFei
Voice input for the web UI: tap Alt (or Alt+Space) to start/stop dictation, browser Web Speech (zero-config) or OpenAI-compatible cloud ASR, optional polish through DSH-configured LLM, settings UI.
Installed cleanly when we ran it
1624318455
Reads assistant replies aloud via free Edge TTS or your own RVC voice models: read-aloud buttons + auto-read, adaptive chunked progressive playback (gapless long reads), one-click voice-pack installs from a registry, and a portable RVC runtime.
Installed cleanly when we ran it
Alan2Z
Zero-dependency, event-driven voice announcement plugin: no extra model, no token cost. Speaks with the system's built-in natural voice, supporting both Windows and macOS; final-reply announcements, approval & question alerts, optional event announcements (turn end, command done, goal change, tool errors, todo updates), replayable final replies, and a bilingual visual settings page.
Installed cleanly when we ran it
CAOGGL
Notifies you when a conversation finishes: plays a sound and shows a Windows notification when the agent goes idle (configurable sound file, volume, debounce/throttle).
Installed cleanly when we ran it
AI-Galaxy-GPU
Per-event sound notifications: turn completion, approval, question, plan-review, goal-blocked, and task-failure each get their own sound and volume, configurable in the Web UI (built-in synth, mute, or local audio file).
Installed cleanly when we ran it
