dsh-tts-supertonic is a plugin for deepseek-harness that converts assistant replies
into speech using the Supertonic 3 ONNX model. Text is never sent to the cloud; everything is processed locally on your device.
Requirements
- Python >=3.13, 64-bit
- Python packages
supertonic,numpy,onnxruntime - Supertonic 3 model, ~400 MB (downloaded automatically on first run)
Windows
install Python from ((https://www.python.org) and tick "Add python.exe to PATH" in the installer.
Do not use the Microsoft Store Python. If you already have it, disable it under Settings -> Apps -> Advanced app settings -> App execution aliases, or set pythonPath to a real interpreter (see Configuration).
python --version
python -m pip install supertonic numpy onnxruntime
python -c "import supertonic, numpy, onnxruntime; print('ok')"
Debian GNU/Linux
Since Debian 12, pip install refuses to install packages into the system Python environment (PEP 668, externally managed environment). Therefore, use a virtual environment as shown below.
If absolutely necessary, you can bypass this restriction using:
pip install supertonic numpy onnxruntime --break-system-packages
Note: Using --break-system-packages is not recommended, as it may interfere with system-managed Python package
Recommended approach:
sudo apt update
sudo apt install python3 python3-venv
python3 -m venv ~/.dsh-tts-venv
~/.dsh-tts-venv/bin/python -m pip install supertonic numpy onnxruntime
~/.dsh-tts-venv/bin/python -c "import supertonic, numpy, onnxruntime; print('ok')"
If the self-test prints ok, the environment is ready.
On Linux, configure the plugin to use the virtual environment's Python interpreter by setting pythonPath in cordis.patch.yml:
dsh-tts-supertonic:
pythonPath: "/home/<user>/.dsh-tts-venv/bin/python"
Configuration
All options are set in cordis.patch.yml:
| Option | Type | Default | Description |
pythonPath |
string | python3 (python on Windows) |
Path to the Python interpreter. Set a full path if python does not resolve to a real interpreter. |
runtimePath |
string | <plugin>/runtime/worker.py |
Path to worker.py. An empty value uses the bundled file. |
modelPath |
string | '' |
Path to a local Supertonic 3 model. An empty value triggers an automatic download. |
voice |
string | F1 |
Voice. |
language |
string | hr |
Speech language (ISO code, e.g. hr, en, de). Supertonic 3 supports 31 languages. |
enabled |
boolean | true |
If false, routes return status 403 and the worker is not started. |
timeoutMs |
number | 120000 |
Maximum duration of a single synthesis request. |
startupTimeoutMs |
number | 180000 |
Maximum time to wait for the worker to become ready. |
maxTextLength |
number | 5000 |
Maximum text length per request, in characters. |
Model and cache
The Supertonic 3 model is not included with the plugin. It is downloaded automatically on first run (~400 MB) and stored in the user cache directory:
- Linux:
~/.cache/supertonic3/ - Windows:
C:\Users\<user>\.cache\supertonic3\
If the first startup times out while the model is still downloading, raise startupTimeoutMs (e.g. 600000).
No comments yet. Be the first to write one.