Staged Wake Word Detection Using Low Power and High Performance Domains
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for processing wake words in smart devices are limited by high power consumption and insecure data transmission, particularly in cloud-based verification operations.
Innovation Solution
The proposed solution involves splitting audio signal detection components between a low power domain and a high performance domain to reduce power consumption, and configuring audio data buffers to operate in either a low power management profile or a performance profile, which can be user-selectable.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If cloud-based verification operations are used for wake word detection, then verification accuracy is improved, but power consumption increases and security is compromised
Solution Approach 1:
The wake word detection system is segmented into two parts: a lightweight model that operates locally on the device for initial detection, and a more complex cloud-based model for verification only when needed. This segmentation allows the device to maintain high accuracy while minimizing power consumption by keeping the simple local model running continuously and using the power-intensive cloud verification only occasionally.
2Measurement precision
If cloud-based verification operations are used for wake word detection, then verification accuracy is improved, but data transmission security is compromised
Solution Approach 1:
The critical verification function is extracted from the cloud and implemented as a lightweight model that runs locally on the device. This extraction eliminates the need to transmit audio data to the cloud for verification, thereby maintaining high verification accuracy while removing the security risk associated with data transmission.
3Measurement precision
If high performance components are used for wake word detection, then detection accuracy is improved, but power consumption increases
Solution Approach 1:
The system dynamically adjusts its operational mode based on conditions. The lightweight model operates continuously in a low-power state for basic detection. When wake words are detected or specific conditions are met, the system dynamically transitions to using the more accurate but power-intensive cloud-based verification model, thereby optimizing the balance between accuracy and power consumption.
Data Source
AI summary
Systems, methods, and devices detect audio signals. Methods may include receiving an audio input at an audio front end circuit, starting, using a low power circuit, one or more buffers in response to receiving the audio input, and identifying, using the low power circuit, speech included in the audio input. Methods may also include identifying, using the low power circuit and a high performance circuit, a wake word based, at least in part, on the identified speech, the high performance circuit being configured to verify the wake word identified by the low power circuit.