Multi-Wakeword Audio Detection With Shared Encoder Decoders
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing availability of multiple wakewords and assistant voice services in devices strains computing resources due to the need for multiple wakeword detectors, making it impractical for devices with limited resources to operate simultaneously.
Innovation Solution
A two-layered approach is implemented, where a joint encoder processes input audio into feature vectors, and individual decoders detect specific wakewords, allowing devices to activate and deactivate decoders based on context to manage computing resources efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple wakeword detectors are implemented to support multiple assistant voice services, then the device can detect multiple wakewords, but the computing resources are overwhelmed
Solution Approach 1:
The wakeword detection system is segmented into two distinct layers: a shared encoder that processes audio input and generates feature vectors, and multiple specialized decoders that each detect specific wakewords. This segmentation allows the computationally intensive encoding to be performed once, while the lighter decoding operations can be selectively executed for multiple wakewords, thereby supporting multiple assistant voice services without overwhelming device resources.
2Device complexity
If a joint encoder is shared across multiple decoders, then computational overhead is reduced, but the system must manage activation and deactivation of decoders based on context
Solution Approach 1:
The system implements dynamic management of decoders based on operational context. Decoders can be selectively activated or deactivated depending on which assistant voice services are currently active or relevant. This dynamic approach allows the shared encoder to efficiently process audio input while only the necessary decoders are engaged, reducing computational overhead and simplifying operational management.
Data Source
AI summary
A device is configured to detect multiple different wakewords. A device may operate a joint encoder that operates on audio data to determine encoded audio data. The device may operate multiple different decoders which process the encoded audio data to determine if a wakeword is detected. Each decoder may correspond to a different wakeword. The decoders may use fewer computing resources than the joint encoder, allowing for the device to more easily perform multiple wakeword processing. Enabling/disabling wakeword(s) may involve the reconfiguring of a wakeword detector to add/remove data for respective decoder(s). Specific decoders may be activated/deactivated depending on device context, thereby efficiently managing device resources.