Show HN: Hushkey – Silent dictation hotkey for macOS (github.com)

A new macOS application called Hushkey has been introduced, allowing users to perform silent dictation by mouthing words without making any sound. By utilizing webcam footage, Hushkey reads lip movements and translates them into text in real-time across various applications. This innovative approach to silent speech recognition represents a significant leap forward for the AI/ML community, showcasing the potential for lip-reading technology to function independently of audio inputs, thus ensuring user privacy and expanding accessibility for those who may be unable to speak vocally. The Hushkey application operates as a local service, processing data entirely on the user’s device without sending video off the machine, which addresses privacy concerns common with similar technologies. Employing a pipeline that combines Apple Vision facial landmarks and advanced lip reading algorithms—along with personalized training for improved accuracy—Hushkey leverages a framework that integrates machine learning components like Auto-AVSR and LoRA adapters. This end-to-end local processing not only enhances dictation accuracy through personalization but also minimizes reliance on external servers, making this a promising tool for those seeking a discreet method of dictation and interaction with technology.