Partial Hotword Detection for Speech Recognition Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems experience latency when waiting for a full hotword to be spoken before performing associated actions, as they typically require the entire hotword to be detected before initiating any actions, leading to delayed processing and user-perceived slowness.
Innovation Solution
Implementing a system that detects an initial portion of a hotword and performs a subset of associated actions immediately, such as waking up a mobile device from sleep or establishing a network connection, allowing for reduced latency and improved processing speed by initiating actions transparently in the background.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the system waits for the entire hotword to be detected before performing actions, then the accuracy of hotword recognition is improved, but the processing time increases
Solution Approach 1:
The system performs preliminary actions by detecting and responding to an initial portion of the hotword before the entire hotword is spoken. This allows the device to wake from sleep mode and prepare processing resources in advance, reducing the overall processing time while maintaining accurate recognition of the complete hotword sequence.
Solution Approach 2:
The hotword detection process is segmented into multiple stages: initial portion detection triggers wake-up and preparation actions, while subsequent portions continue to be processed for complete hotword confirmation. This segmentation allows parallel execution of wake-up actions and ongoing speech processing, reducing latency.
2Reliability
If the system performs actions only after detecting the complete hotword, then the reliability of action execution is improved, but the responsiveness to user input deteriorates
Solution Approach 1:
The system executes preliminary actions upon detecting the initial portion of the hotword, such as waking from sleep mode and initializing processing resources. These preliminary actions improve responsiveness without compromising the reliability of complete hotword recognition, as the system continues to process the full hotword sequence for final action confirmation.
3Speed
If the system activates all actions upon detecting any part of the hotword, then the responsiveness is improved, but the energy consumption increases
Solution Approach 1:
The system segments action execution into preliminary actions and main actions. Preliminary actions (e.g., waking from sleep mode) are triggered by the initial hotword portion, while main actions are executed only after complete hotword recognition. This segmentation improves responsiveness for critical wake-up functions while conserving energy by delaying resource-intensive actions until necessary.
Solution Approach 2:
The system performs a partial set of actions (prelimiminary wake-up actions) upon detecting the initial hotword portion, rather than executing the complete action set. This partial action approach improves perceived responsiveness while minimizing energy consumption by activating only essential functions during the interim period.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, receiving audio data; determining that an initial portion of the audio data corresponds to an initial portion of a hotword; in response to determining that the initial portion of the audio data corresponds to the initial portion of the hotword, selecting, from among a set of one or more actions that are performed when the entire hotword is detected, a subset of the one or more actions; and causing one or more actions of the subset to be performed.