Lightweight Custom Wake Words Project

🧑‍💻 Objective Voice interaction or commands “Hey Google”, or “Hey Siri” rely on keyword spotting to start interaction on local device. It helps people experience “Hands-free” searching and task completion

💭 What & Why - Challenges in above technology

• Triggers on negative wake words, unrelated speech, background noise, or silence
• High no. of instance, when device does not trigger on positive wake words
• Need for quick response & acknowledgment
• Ability to customise wake word
• Wake model to be lightweight & energy efficient

🔬 Approach To design, build and deploy a lightweight Keyword spotting ML model (CNN, SVM) and exposed as a mobile-application that can process a “custom wake word”. Voice response with results by respecting local device resource constraints (low compute) and adhering to ethical challenges (Privacy respecting and non-eavesdropping)

🧑‍💻 Github Link