DSH HUB
HomePlugin StorePlugin PacksCommunityRankingsResourcesPublish Guide
Plugin source
Back to catalog

tp5s5mvjpv-coder /

tp5s5mvjpv-coder/dsh-voice-endpoint

Topic repository only

Pause-terminated voice input for DeepSeek Harness: speak, pause, and the take transcribes itself.

★ 0 Stars0 Forks0 IssuesN/A Community rating0 Confirmed installs
View on GitHub
READMESource: main@707e9208

dsh-voice-endpoint

Voice input for DeepSeek Harness that ends the take when you pause, instead of waiting for a Stop click.

Click the microphone once, speak, stop talking. After about a second of silence the take ends, the configured speech provider transcribes it, and the text lands in the composer draft and is submitted for you.

对话输入框:[麦克风] 你说的话……(停顿 1s)
          → 自动结束录音 → 识别中… → 文字插入草稿 → 自动发送 →(麦克风自动重开,直接说下一句)

Packages

Package Role
dsh-voice-endpoint Client plugin: patched toolbar that ends a take on a detected pause
dsh-voice-endpoint-bundle Profile bundle: disables the stock ui-voice-input row and inserts the plugin

Install only the bundle; it depends on the plugin.

Install

dsh plugin --profile web add dsh-voice-endpoint-bundle

Desktop builds manage the reserved desktop profile:

dsh plugin --profile desktop add dsh-voice-endpoint-bundle

Then restart DeepSeek Harness.

Uninstall restores the stock toolbar:

dsh plugin --profile web remove dsh-voice-endpoint-bundle

Tuning

Constants at the top of packages/dsh-voice-endpoint/lib/client.js:

const ENDPOINT_RMS = 0.012        // raise in a noisy room
const ENDPOINT_POLL_MS = 60
const ENDPOINT_MIN_SPEECH_MS = 200
const ENDPOINT_SILENCE_MS = 1000  // silence that ends the take

docs/patch.md holds the 39-line unified diff against upstream src/client/VoiceInput.tsx, for the day the feature lands in tree.

How it works

The stock plugin already samples the microphone to draw its waveform. This build reuses that reading: speech crossing an RMS gate accumulates speaking time, then continuous silence calls the same finish() the Stop button calls. No provider change: SenseVoice already transcribes a complete clip and does its own VAD segmentation.

Compatibility

Built against DSH 0.2.1-rc.3. The bundle patch targets the stock row id ui-voice-input. If a future DSH release renames that row or changes the client plugin contract, the patch cannot apply and the microphone disappears — uninstall to recover. This plugin does not add streaming transcripts.

License

MIT. packages/dsh-voice-endpoint/lib/client.js is a modified copy of @deepseek-ai/dsh-experimental-client-ui-voice-input (MIT, DeepSeek).

—/ 5

No ratings yet

Manifest verification required

Commit 707e9208ca41

Community comments

No comments yet. Be the first to write one.

DSH HUB

A community index for DSH plugins. Not an official GitHub or DeepSeek AI product.

CommunityResourcesAPIAbout