Multilingual Keyword Detection via Decoding Network Penalty
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing keyword detection systems in speech recognition face inefficiencies when dealing with multiple languages, particularly due to the need for a prolonged language recognition phase and inaccuracies when keywords from different languages are mixed, leading to suboptimal keyword detection performance.
Innovation Solution
A method utilizing a decoding network that includes keywords and filler words from multiple languages, where a confidence score is updated with a penalty factor when distinct languages are detected, allowing for efficient and accurate keyword recognition without requiring a separate language recognition phase.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a separate language recognition phase is performed before keyword detection, then language identification accuracy is improved, but processing time and system complexity increase
Solution Approach 1:
The patent merges the language recognition phase and keyword detection phase into a single integrated processing stage. The decoding network simultaneously performs both language identification and keyword detection without requiring a separate language recognition phase, thereby reducing processing time while maintaining accuracy through the unified model's coordinated operation of language modeling and keyword detection components
Solution Approach 2:
The decoding network is designed as a universal system that performs multiple functions simultaneously: it recognizes languages, detects keywords, and processes speech data all within the same computational framework. This multi-functional approach eliminates the need for separate dedicated phases for language recognition and keyword detection, resolving the time complexity issue
2Measurement precision
If extended speech data is used for language recognition, then language identification accuracy is improved, but streaming keyword detection capability deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-processing speech data into discrete frames and using a decoding network that is prepared to process each frame independently while maintaining context awareness. The system performs language recognition and keyword detection on each frame as it becomes available, rather than waiting for extended speech segments, enabling real-time streaming processing with maintained accuracy through the pre-configured decoding model
Solution Approach 2:
The system dynamically adjusts its processing approach by continuously analyzing each speech frame and updating language and keyword detections in real-time. The decoding network adapts to varying speech patterns and language switches on-the-fly, allowing streaming keyword detection to maintain both speed and accuracy without requiring extended pre-processing of the entire speech stream
3Adaptability or versatility
If keywords from multiple languages are mixed in one sentence, then language diversity is improved, but keyword detection accuracy deteriorates
Solution Approach 1:
The patent applies local quality by making the decoding network sensitive to local language context changes. The system identifies and adapts to language switches at specific positions in the speech stream, adjusting its keyword detection parameters locally for each language segment. This allows accurate keyword detection in mixed-language sentences by treating different language regions with appropriate detection parameters rather than applying a uniform approach throughout
Solution Approach 2:
The decoding network incorporates feedback mechanisms that continuously monitor language identification results and adjust keyword detection accordingly. When language switches are detected, the system receives feedback about the current language context and modifies its keyword detection strategy in real-time, ensuring high accuracy even when multiple languages are mixed in the same sentence
Data Source
AI summary
This application discloses a method implemented of recognizing a keyword in a speech that includes a sequence of audio frames further including a current frame and a subsequent frame. A candidate keyword is determined for the current frame using a decoding network that includes keywords and filler words of multiple languages, and used to determine a confidence score for the audio frame sequence. A word option is also determined for the subsequent frame based on the decoding network, and when the candidate keyword and the word option are associated with two distinct types of languages, the confidence score of the audio frame sequence is updated at least based on a penalty factor associated with the two distinct types of languages. The audio frame sequence is then determined to include both the candidate keyword and the word option by evaluating the updated confidence score according to a keyword determination criterion.