Server-Side Hotword Detection With Two-Threshold Audio Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic speech recognition systems face challenges in efficiently identifying hotwords with high accuracy and reduced resource usage, particularly in client devices, leading to increased processing time and bandwidth consumption.
Innovation Solution
Implementing a two-tiered threshold system where client devices use a lower threshold for initial hotword detection and then send audio signals to a server system using a more restrictive threshold for confirmation, leveraging client-side and server-side models trained on different data volumes to enhance accuracy and reduce false positives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a more restrictive threshold is used for hotword detection, then hotword detection accuracy is improved, but processing time and resource consumption increase
Solution Approach 1:
The patent segments the hotword detection process into two stages: a client-side stage using a lower threshold for quick initial filtering, and a server-side stage using a more restrictive threshold for accurate confirmation. This segmentation allows the system to achieve high accuracy without the full processing time cost of always using the restrictive threshold.
Solution Approach 2:
The patent introduces an intermediary approach where the client device performs preliminary analysis with a lower threshold and only sends audio signals to the server for confirmation when the client's initial detection meets the lower threshold. This intermediary step reduces unnecessary server processing and bandwidth usage while maintaining accuracy.
2Reliability
If a more restrictive threshold is used for hotword detection, then false positives are reduced, but bandwidth consumption increases
Solution Approach 1:
The client device performs preliminary hotword detection using a lower threshold before sending audio signals to the server. This preliminary action filters out obvious non-hotwords locally, reducing the number of signals that need to be transmitted to the server and thus reducing bandwidth consumption while maintaining reliability.
Solution Approach 2:
The patent extracts the initial hotword detection function from the server and places it on the client device. By taking out the detection logic to the client, the system reduces the volume of data that needs to be transmitted to the server, thereby reducing bandwidth consumption while the server maintains final confirmation authority for accuracy.
3Measurement precision
If client devices perform comprehensive hotword analysis, then detection accuracy is improved, but device resources are consumed
Solution Approach 1:
The client device performs partial hotword analysis using a lower threshold rather than comprehensive analysis with the more restrictive threshold. This partial action is sufficient for initial filtering and reduces client resource consumption, while the server completes the comprehensive analysis when needed to maintain accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for detecting hotwords using a server. One of the methods includes receiving an audio signal encoding one or more utterances including a first utterance; determining whether at least a portion of the first utterance satisfies a first threshold of being at least a portion of a key phrase; in response to determining that at least the portion of the first utterance satisfies the first threshold of being at least a portion of a key phrase, sending the audio signal to a server system that determines whether the first utterance satisfies a second threshold of being the key phrase, the second threshold being more restrictive than the first threshold; and receiving tagged text data representing the one or more utterances encoded in the audio signal when the server system determines that the first utterance satisfies the second threshold.