RelayTive

RelayTive is a privacy-first iOS app for translating atypical speech into clear, typical language.
All processing is local using CoreML and HuBERT for on-device audio recognition.

Key Features

• Train custom mappings between unique vocalizations and their explanations.
• Real-time translation using user-created training data.
• All data and processing stay on your device.

Installation

1. Clone this repo and open RelayTive.xcodeproj in Xcode 15 or newer.
2. Add your HuBERT CoreML model (RelayTive_HuBERT.mlpackage) to RelayTive/Models/.
3. Set up microphone and speech recognition permissions in Info.plist.
4. Build and run on a physical iOS device.

Privacy

RelayTive does not collect or transmit user data.
All recognition and storage are performed locally.

Note

The HuBERT model file is not included in this repository.

Model provenance

We use HuBERT Base (LS-960) for frame-level embeddings.

• Checkpoint: hubert_base_ls960.pt
• SHA-256: 1703cf8d2cdc76f8c046f5f6a9bcd224e0e6caf4744cad1a1f4199c32cac8c8d
• Size: 1,136,468,879 bytes
• Source:
• License (weights): <e.g., CC BY-NC 4.0> — demo/research only if non-commercial.