utterance-core 0.0.1

Client-side semantic endpointing. Know when they're done talking.

Utterance Python SDK

Client-side semantic endpointing. Know when they're done talking.

This is the Python SDK for the Utterance model. It provides a simple interface to load the ONNX model and run inference for VAD/endpointing.

Installation

pip install utterance

(Note: This package is not yet published to PyPI. Install locally or use pip install -e .)

Usage

from utterance import Utterance
import numpy as np

# Initialize detector (downloads model automatically)
detector = Utterance()

# Create dummy features (replace with real MFCCs/features)
# Example shape: (1, 100, 40) - batch, time, features
dummy_features = np.random.randn(1, 100, 40).astype(np.float32)

# Run inference
result = detector.predict(dummy_features)

print("Speaking:", result["speaking"])
print("Thinking Pause:", result["thinking_pause"])
print("Turn Complete:", result["turn_complete"])

Contributing

See main repository CONTRIBUTING.md.