v1.6.0: Optional LSTM Neural Network Predictor

What's New

• Optional LSTM neural network predictor trained on 116 Project Gutenberg books (20,000-word vocabulary, 21MB ONNX model). The LSTM provides context-aware predictions that complement the n-gram model. Toggle in Settings > Word Predictor.
• Bundled onnxruntime DLLs (v1.20.1) in lib/ for ctypes fallback when the Python onnxruntime package is not installed.
• The n-gram model remains the default. The LSTM is opt-in.

How to Use the LSTM

1. Install the add-on and restart NVDA.
2. Go to NVDA Menu > Preferences > Settings > Word Predictor.
3. Check "Use LSTM neural network predictor."
4. Click OK. Predictions now come from the neural network.

If onnxruntime is not installed via pip, the bundled DLLs are used automatically.

Download

Download the .nvda-addon file below and open it to install.