Transcribro offers a private, on-device speech-to-text solution for Android, functioning as both a standalone keyboard and a system service. It targets Android users prioritizing data privacy and local processing for voice input, providing a secure alternative to cloud-based services.

How It Works

The system leverages whisper.cpp for efficient, on-device execution of OpenAI's Whisper models, enabling local speech recognition. Voice Activity Detection (VAD) is handled by Silero VAD, ensuring accurate segmentation of speech. This combination allows for real-time transcription directly on the device without sending audio data to external servers.

Quick Start & Requirements

• Install: Download from Accrescent app store (recommended) or GitHub releases.
• Prerequisites: Android device. APK verification via apksigner is recommended if downloading directly.
• Resources: No specific hardware requirements beyond a functional Android device are listed, implying compatibility with a wide range of hardware.
• Docs: https://github.com/soupslurpr/Transcribro

Highlighted Details

• On-device processing for enhanced privacy.
• Utilizes whisper.cpp for efficient model execution.
• Functions as a system-wide speech-to-text input method.

Maintenance & Community

• Community support is available via a Matrix space: https://matrix.to/#/#transcribro:matrix.org
• Contribution guidelines are provided in CONTRIBUTING.md.

Licensing & Compatibility

• The README does not explicitly state a license. Branding restrictions are in place for derivative works.

Limitations & Caveats

Currently, Transcribro only supports English language recognition, although multi-language support is planned. The branding restrictions may impact the creation of modified versions intended for public release.