Software Install (Version 1)

Before you start: create your Nova folder

Put everything inside one folder in your Documents, like: C:\Users\YourName\Documents\Nova Desktop\

You can create this folder in Windows (right-click → New folder) or using Command Prompt:

cd %USERPROFILE%\Documents
mkdir "Nova Desktop"

Your Nova folder should contain: nova_cli.py, face_tracking_0.3.py, and an animations folder. (These come from the download pack.)

1) Install Python

Install Python 3.10 (64-bit) (recommended for best compatibility).

During install, tick: Add Python to PATH

Verify in Command Prompt:

python --version
pip --version

If python doesn’t work, try py instead:

py -3.10 --version

2) Install required libraries

These cover nova_cli (speech + LLM + audio) and face tracking (camera + tracking + UDP).

Core (Nova CLI)

pip install numpy sounddevice simpleaudio ollama faster-whisper pygrabber

Face tracking (RealSense + tracking)

pip install opencv-python mediapipe pyrealsense2

Common extras

pip install requests

If you’re using a RealSense camera, you may also need Intel’s RealSense drivers installed on Windows.

3) Install / run Ollama

Install Ollama for Windows and confirm it runs.

Download Ollama (Windows)

Requires Windows 10 or later.

ollama --version
ollama pull nova-phi4-mini
ollama list
ollama run nova-phi4-mini

Verify the model name in your script matches what you pulled (example: nova-phi4-mini). If your model name is different, update MODEL_NAME in your Python scripts.

Nova Modelfile (required for “Nova” behaviour + animations)

Nova uses a custom Modelfile so Ollama responds in Nova’s format (including animation/emotion tags). You’ll find the Modelfile in your download pack.

Copy the file named Modelfile into your Windows home folder (your user directory): C:\Users\YourName\Modelfile (same as %USERPROFILE%\Modelfile).

Then create the Nova model from that file:

cd %USERPROFILE%
ollama create nova-phi4-mini -f Modelfile

After that, ollama run nova-phi4-mini should behave like Nova, and your scripts can use MODEL_NAME = "nova-phi4-mini" for the correct responses.

Modelfile errors? Make sure the file is named exactly Modelfile (no .txt), and re-run the create command.

4) Install Piper (TTS) (Windows)

Piper is Nova’s text-to-speech engine. nova_cli.py expects Piper in these exact locations by default:

Piper executable: C:\piper\piper.exe
Voice models folder: C:\piper\models\

This means you don’t need to add Piper to PATH — the script calls it by full path.

A) Create the Piper folders (Command Prompt)

mkdir C:\piper
mkdir C:\piper\models

B) Download Piper (zip) + extract to C:\piper\

Download the Windows release zip from the official Piper releases page and extract it into: C:\piper\

Open Piper releases (GitHub)

After extracting, you should have piper.exe directly inside C:\piper\. If the zip extracts an extra folder layer, move the contents so that piper.exe sits directly in C:\piper\.

C) Test Piper (Command Prompt)

C:\piper\piper.exe --help

D) Download a voice model (put it in C:\piper\models\)

Piper needs a voice. Download both files for a voice: .onnx and the matching .onnx.json, and put them in: C:\piper\models\

Browse Piper voices

Example: if you choose en_GB-cori-high, you’ll typically download: en_GB-cori-high.onnx and en_GB-cori-high.onnx.json and place both into C:\piper\models\.

E) Generate a test WAV

Replace the voice filename with the one you downloaded:

echo Hello, I am Nova. | C:\piper\piper.exe --model "C:\piper\models\en_GB-cori-high.onnx" --output_file "C:\piper\test.wav"

F) Confirm nova_cli.py matches your chosen voice

In nova_cli.py, the Piper command looks like this. Make sure the --model path matches the voice you picked:

command = (
    f'echo {safe_text} | '
    f'"C:\\piper\\piper.exe" '
    f'--model "C:\\piper\\models\\en_GB-cori-high.onnx" '
    f'--output_file "{wav_path}"'
)

If you picked a different voice, change en_GB-cori-high.onnx to your chosen filename. Also make sure the matching .onnx.json file is in the same folder as the .onnx.

Piper errors? Confirm C:\piper\piper.exe exists and your voice .onnx + .onnx.json are inside C:\piper\models\.

5) Audio setup (mic + speaker)

Plug in your mic and speakers/headphones.
In Windows Sound settings, set the correct Input device (mic).
If Whisper cuts out early, increase mic gain and speak closer to the mic.
If sounddevice complains about devices, reinstall it and reboot. Audio drivers can be fussy on Windows.

6) RealSense + face tracking setup (Version 1)

Connect the RealSense via USB 3.0 (blue port if possible).
Run your face tracking script and confirm the camera feed appears.
If it can’t find the camera, install RealSense drivers and try another USB port.

Your nova_cli.py launches face tracking automatically. If you moved files/folders, update: FACE_TRACKING_SCRIPT = r"...\\face_tracking_0.3.py"

7) Networking / UDP (Nova servo boards)

Your computer and ESP32 boards must be on the same Wi-Fi network.
To find your PC IP: open Command Prompt and run:

ipconfig

If you need to change board IPs, update these in nova_cli.py: ARM_IP, TORSO_IP, and ports.

When the ESP32 successfully connects to your router, the little white status LED will flash on and off.

8) Run Nova

From your Nova folder:

cd "%USERPROFILE%\Documents\Nova Desktop"
python nova_cli.py

Your Nova folder should contain: nova_cli.py, face_tracking_0.3.py, and an animations folder.

If you stored Nova somewhere else, update the paths inside nova_cli.py, especially: ANIM_DIR and FACE_TRACKING_SCRIPT.

Trouble?

Module not found: run the pip install ... commands again, then retry.
RealSense not detected: try a different USB port + install RealSense drivers.
No audio input: check Windows input device + mic permissions.
Ollama errors: verify Ollama is running and the model name exists.
Modelfile errors: file must be named exactly Modelfile (no .txt), then run ollama create ... again.
Piper errors: confirm C:\piper\piper.exe exists and voice files are in C:\piper\models\.

Contact