PolinniZhong/dsh-omi-voice

中文 | English

Hear AI replies in DeepSeek Harness (DSH) spoken aloud in a natural Doubao TTS voice. Voice synthesis runs in a local Omi engine on your Mac, and your Doubao API Key stays in your own Keychain (BYOK) — the plugin never touches your key.

Overview

Item	Description
Name	dsh-omi-voice
Platform	DeepSeek Harness (dsh web, incl. desktop) + macOS Omi engine v0.1.3+
Problem solved	DSH's built-in TTS (system voice / Edge TTS) sounds robotic; clipboard tools force you to copy first
How it works	Tap 🔊 on a reply to read / pause / resume; reads only the final answer — tool logs, code, tables, diagrams are filtered out before synthesis
Voice	Doubao TTS 1.0 (seed-tts-1.0), natural Chinese voice, pause/resume from where you paused
Privacy	API Key lives in the Omi engine's macOS Keychain; plugin has zero key access; localhost-only (127.0.0.1)
Cost	BYOK, billed by character; in-memory LRU cache + dedup avoid repeat billing
License	MIT

Get started in 3 steps

1. Install the plugin and build/open the Omi engine (see "Getting a Doubao API Key" and "Installation" below).
2. Save your Doubao API Key once in the Omi engine settings.
3. Tap 🔊 next to any AI reply in DSH to hear it.

flowchart LR
    A[Tap 🔊] --> B[Plugin extracts the final answer text]
    B --> C[POST 127.0.0.1:8765/v1/speak]
    C --> D[Omi engine cleans + segments]
    D --> E[Doubao TTS streaming synthesis]
    E --> F[Local speaker playback]

Getting a Doubao API Key (first-time users)

This plugin is BYOK: Doubao speech is billed to your own Volcengine account by character count. Three steps:

Step 1 · Find "Doubao Speech": log in to the Volcengine Console and open the "Doubao Speech" (speech technology) service.

Step 2 · Enable "Speech Synthesis 1.0": enable the "Speech Synthesis 1.0" product in the service list.

Step 3 · Create an API Key: when creating an Access Key, link it to the "Speech Synthesis 1.0" service you enabled; paste the resulting API Key into the Omi engine settings ("Settings > API Key") and save.

The API Key is stored only in the Omi engine's macOS Keychain — the plugin has zero key access and nothing leaves your machine.

Installation

dsh plugin --profile web add "github:PolinniZhong/dsh-omi-voice#v0.1.3&path:/"

For local development you can install a directory directly:

dsh plugin --profile web add /path/to/dsh-omi-voice

Engine (Omi DSH) build steps: see engine/README.md — run ./engine/build/build-service.sh, then ditto the result to ~/Applications/Omi DSH.app.

Usage

• Tap 🔊 to read; tap again while playing = pause; tap again = resume from where you paused; tapping 🔊 on another message interrupts and reads the new one.
• Reads only the final answer: tool execution logs and thinking are not read; code fences, tables, and pure graphics (box-drawing/ASCII) are filtered before the request — replies containing only such content show a disabled button with "nothing to read".
• When the engine is off or the key is missing, tapping shows a clear hint (including where to open Omi).

Cost transparency

Doubao TTS (seed-tts-1.0) is billed by character to your Volcengine account. Therefore:

• Synthesis happens only when you manually tap 🔊 — no auto-read;
• The engine dedups identical text within 3 seconds and keeps an in-memory LRU cache (last 3 items, ≤5MB, memory-only, cleared on quit) so re-reading recent content doesn't re-bill;
• Replies with no speakable content (pure tables/code) never hit the TTS API (invalid_text).

Architecture & protocol

The plugin is just a "remote control": playback, speed, pause/resume, caching, and text cleaning all happen in the Omi engine. Protocol: docs/API.md (/v1/status, /v1/speak, /v1/pause, /v1/resume, /v1/stop).

FAQ

Question	Answer
Tap 🔊 shows "Omi DSH engine not detected"	Open the "Omi DSH" app (~/Applications or press ⌘Space and search "Omi"); consider enabling "Launch at login"
"Configure your Doubao API Key in Omi settings"	Open "Omi DSH" settings and save the key once (see "Getting a Doubao API Key")
Button is greyed out / nothing happens	The reply has no readable content (pure code/table/diagram), auto-filtered
Can I change the voice?	Yes, via the Omi engine settings "Voice ID" (the plugin has no voice picker yet)
Windows support?	Not yet — the Omi engine is macOS Apple Silicon only
vs. dsh-voice-chat	It is free/no-key but uses the robotic system voice; this plugin uses the natural Doubao voice at the cost of BYOK character billing

Related

• Engine: this repo's engine/ (Omi DSH local engine, Swift source + build scripts)
• Ecosystem: awesome-dsh-plugin

Project docs

• AGENTS.md — agent instructions (Codex convention)
• docs/DESIGN.md — architecture & design
• docs/DECISIONS.md — design decision log
• docs/MEMORY.md — long-term knowledge (gotchas & conclusions)
• docs/HANDOFF.md — handoff & continuation
• docs/API.md — local protocol

License

MIT