Introduction¶
DeepSeek Harness uses a plugin-based architecture. For developers who need to run large model inference locally, speech synthesis is a common integration requirement. This plugin aims to connect existing local TTS engines and make them available in the DSH environment.
What This Is¶
This is a local process launcher for IndexTTS 2.5 and GPT-SoVITS. It communicates with local TTS engines via the JSONL protocol, does not include model weights or a Python environment, and is only responsible for launching configured local processes and forwarding synthesis requests.
Core Features¶
- Sentence-level speech synthesis with IndexTTS 2.5
- Sentence-level speech synthesis with GPT-SoVITS
- Local process launcher
- Edge TTS support
- RVC support
- Auto Read toggle
Installation and Enablement¶
Install the plugin using the following command:
dsh plugin --profile web add github:abbccdd/dsh-localtts
After installation, restart dsh web and refresh the browser page to activate the plugin.
Configuration and Usage¶
- Open settings: Go to
Settings → Plugins → Voice · Local AI TTS. - Configure the engine: Select
IndexTTS 2.5orGPT-SoVITSin the engine option. - Set the project folder: Enter the absolute path of the installed environment. The plugin will discover configuration files under this path.
- Set the reference audio: Select the reference audio file required by the model.
- Enable automatic reading: Turn on the
Turn Auto Read ONswitch in the interface.
Use Cases and Notes¶
- Environment dependencies: Model weights and a Python environment are not included; you need to prepare the installation of IndexTTS 2.5 or GPT-SoVITS yourself.
- Runtime environment: A Node.js 22+ environment and the web profile of DeepSeek Harness are required.
- Audio control: This plugin cannot stop audio that is being played by the original plugin.
- License: MIT License.