Keyword Spotter Using Frame-by-Frame Likelihood Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition systems face high latency and reduced accuracy in spotting keywords, as they require processing the entire utterance and assume silence after the keyword, which is not practical for real-time applications.
Innovation Solution
A keyword spotter that continuously updates probabilities for each audio frame to determine keyword likelihood without requiring silence, using a finite state transducer to process frames efficiently and manage traceback data to enable rapid keyword recognition without substantial delay.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the system processes the entire utterance to spot keywords, then keyword recognition accuracy is improved, but latency increases
Solution Approach 1:
The system performs preliminary keyword spotting during the first speech recognition pass before the utterance is complete. By calculating keyword likelihoods frame-by-frame and checking for keyword endings at each frame, the system can identify keywords in real-time without waiting for full utterance processing, thereby reducing latency while maintaining accuracy through subsequent verification passes.
2Loss of time
If the system assumes silence after keyword to reduce processing time, then latency is reduced, but reliability decreases
Solution Approach 1:
The system continuously performs keyword spotting without requiring silence periods. The keyword likelihood calculation and ending detection operate continuously across all audio frames, allowing the system to reliably identify keywords even when followed by additional speech. This continuous processing maintains reliability while reducing the latency associated with waiting for silence.
3Measurement precision
If the system performs multiple speech recognition passes to improve accuracy, then keyword recognition accuracy is improved, but computational complexity increases
Solution Approach 1:
The system extracts keyword spotting functionality from the full speech recognition pipeline and performs it independently during the first pass. By calculating keyword likelihoods separately using extracted audio features and HMM scores, the system identifies keywords without requiring multiple complete speech recognition passes, thereby reducing computational complexity while maintaining accuracy.
4Measurement precision
If the system stores traceback data for all frames to enable accurate keyword spotting, then keyword recognition accuracy is improved, but memory usage increases
Solution Approach 1:
The system applies different data retention strategies to different time periods. Traceback data is stored for recent frames that are relevant to current keyword detection, while older frame data is discarded. This local quality approach maintains sufficient data for accurate keyword spotting in the current time window while reducing overall memory usage by not retaining historical data that is no longer relevant.
Data Source
AI summary
Features are disclosed for spotting keywords in utterance audio data without requiring the entire utterance to first be processed. Likelihoods that a portion of the utterance audio data corresponds to the keyword may be compared to likelihoods that the portion corresponds to background audio (e.g., general speech and/or non-speech sounds). The difference in the likelihoods may be determined, and keyword may be triggered when the difference exceeds a threshold, or shortly thereafter. Traceback information and other data may be stored during the process so that a second speech processing pass may be performed. For efficient management of system memory, traceback information may only be stored for those frames that may encompass a keyword; the traceback information for older frames may be overwritten by traceback information for newer frames.