Voice Control System Using Wake Word Detection to Reduce Power Consumption
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems face issues with battery drain and false triggers in battery-operated devices, as well as inefficiencies in controlling devices like lamps, where they consume unnecessary power and trigger falsely due to extraneous noises or conversations, necessitating a more sophisticated method for controlling device operation by voice commands.
Innovation Solution
A speech recognition system utilizing multiple auxiliary sensing devices to control the power state of the recognizer, including light and acoustic sensors, to minimize false triggers and power consumption, allowing voice-controlled operation only when needed, and enabling wake-up from a low power state without button presses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If the speech recognizer is always on and always listening for commands, then the device can respond to voice commands immediately, but battery life is quickly depleted and false accept rate increases
Solution Approach 1:
The system performs preliminary action by having the speech recognizer continuously monitor for wake words in a low-power state. When a wake word is detected, the system then activates full speech recognition processing. This preliminary monitoring action allows the system to remain energy-efficient while still being ready to respond to voice commands when needed.
Solution Approach 2:
The system dynamically adjusts its operational state based on environmental conditions and user presence. The speech recognizer transitions between different power states (always-on listening, periodic checking, or fully active) depending on factors such as ambient noise levels, detected user presence, and current operational context, thereby optimizing the balance between responsiveness and energy consumption.
2Ease of operation
If the speech recognizer is always on and always listening for commands, then the device can respond to voice commands immediately, but the false accept rate increases due to unintended sounds being misinterpreted
Solution Approach 1:
The system introduces an intermediary wake word detection layer between ambient sounds and full speech recognition processing. The wake word acts as a mediator that filters and validates incoming audio signals before triggering the full recognition system, thereby reducing false accepts from unintended sounds while maintaining quick response to legitimate commands.
Solution Approach 2:
The system performs preliminary validation by continuously monitoring for specific wake words before activating full speech recognition. This preliminary action serves as a filter that distinguishes between legitimate command intent and random ambient noise, thereby reducing false accept rates while keeping the system responsive to actual user commands.
3Use of energy by moving object
If a button press or switch is required to turn on the speech recognizer, then battery life is saved by powering down when not in use, but the device requires use of eyes, hands, and feet to locate and activate it
Solution Approach 1:
The system replaces the mechanical activation method (button press or switch requiring physical interaction) with an acoustic field-based activation method (voice-activated wake word detection). This substitution allows the device to be activated hands-free and eyes-free, improving ease of operation while maintaining battery efficiency through selective power management.
Solution Approach 2:
The system provides self-service activation by automatically detecting wake words from the acoustic environment and triggering its own power-on sequence without requiring external physical interaction. The device monitors for activation commands autonomously and activates itself when the appropriate wake word is detected, thereby eliminating the need for manual button presses or switches.
4Reliability
If a dual triggered or gated approach is used with a trigger word, then false accept rate is decreased by requiring two hurdles, but the false reject rate increases and usage becomes more cumbersome
Solution Approach 1:
The system uses preliminary wake word detection as a natural filter rather than a cumbersome dual-trigger requirement. The wake word monitoring occurs continuously in a low-power state, and when detected, it automatically triggers the full speech recognition system. This preliminary action reduces false accepts while maintaining simplicity, as users only need to speak the wake word naturally rather than recall and speak a series of trigger words.
Data Source
AI summary
The present invention includes a speech recognition system comprising a light element, a power control switch, the power control switch varying the power delivered to the light element, a controller, a microphone, a speech recognizer coupled to the microphone for recognizing speech input signals and transmitting recognition results to the controller, and a speech synthesizer coupled to the controller for generating synthesized speech, wherein the controller varies the power to the light element in accordance with the recognition results received from the speech recognizer. Embodiments of the invention may alternatively include a low power wake up circuit. In another embodiment, the present invention is a method of controlling a device by voice commands.