User-Dependent Key Phrase Detection via Sequence Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing key phrase detection systems are limited by their user independence and high error rates, particularly in noisy conditions and when users have accents not recognized by traditional approaches, leading to increased false accepts and rejects.
Innovation Solution
A user-dependent key phrase detection system is implemented, where users can define their own key phrases by enrolling them through audio input, using a deep neural network to generate a key phrase recognition model that includes a single rejection state and multiple key phrase models, allowing for flexible and accurate detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If user independent key phrase detection is implemented, then system simplicity is maintained, but false accept and reject rates increase
Solution Approach 1:
The system performs preliminary enrollment actions where user-specific audio samples are collected and processed to create personalized key phrase models before actual detection begins. This preliminary customization enables the system to adapt to individual user characteristics, thereby reducing false accepts and rejects while maintaining operational simplicity.
Solution Approach 2:
The system changes the detection parameters dynamically by switching between user-independent and user-dependent models based on enrollment status. When users complete enrollment, the system adjusts its detection sensitivity and acoustic model parameters to match that specific user's speech characteristics, improving reliability without permanently increasing complexity.
2Ease of manufacture
If traditional key phrase detection is used, then implementation ease is maintained, but error rates remain high in noisy conditions and with accented speech
Solution Approach 1:
The system implements dynamic adaptability by allowing users to enroll their specific speech patterns and having the detection model adjust accordingly. This dynamic customization enables the system to handle noisy conditions and accented speech effectively, improving measurement precision while maintaining ease of implementation through an automated enrollment process.
Solution Approach 2:
The system provides self-service through automated enrollment where users simply speak the key phrase and the system automatically processes their audio samples to create personalized detection models. This eliminates the need for complex manual configuration while achieving high detection accuracy tailored to each user's speech characteristics.
3Measurement precision
If user dependent key phrase modeling is implemented, then detection accuracy improves, but system complexity increases
Solution Approach 1:
The system segments the detection process into distinct phases: enrollment phase where user-specific models are created, and detection phase where these models are applied. This segmentation allows the complex user-dependent modeling to occur only during the initial enrollment phase, while the ongoing detection operation remains relatively simple, thus improving accuracy without permanently increasing system complexity.
Solution Approach 2:
The complex user-dependent model training is performed as a preliminary action during enrollment, creating ready-to-use detection models in advance. This shifts the computational complexity from the ongoing detection process to the initial setup phase, allowing high detection accuracy to be achieved while keeping the operational system relatively simple.
Data Source
AI summary
Techniques related to key phrase detection for applications such as wake on voice are discussed. Such techniques may include determining a sequence of audio units for received audio input representing a user defined key phrase, eliminating audio units from the sequence to generate a final sequence of audio units, and generating a key phrase recognition model representing the user defined key phrase based on the final sequence.