AI Wakeup Word Detection Using Syllable Sequences
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition technologies for wake-up words are inflexible, requiring retraining of deep neural networks when wake-up words are changed, limiting user customization and increasing computation complexity.
Innovation Solution
An AI-based wake-up word detection method using a preset pronunciation dictionary to construct syllable combination sequences for self-defined wake-up words, extracting speech features, and inputting them into a pre-constructed DNN model to calculate confidence, allowing for fast and efficient detection without the need for special optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a fixed wakeup word solution using DNN acoustic model is used, then the speech recognition accuracy is improved, but the adaptability for user customization deteriorates
Solution Approach 1:
The wakeup word detection is segmented into two independent parts: a fixed DNN acoustic model for speech feature recognition and a flexible syllable combination sequence matcher for custom wakeup word detection. This segmentation allows the acoustic model to maintain high accuracy while the syllable combination approach enables easy user customization without retraining.
Solution Approach 2:
A syllable combination sequence constructed from a pronunciation dictionary serves as an intermediary between the fixed DNN acoustic model and the customizable wakeup words. The acoustic model outputs syllable-level probabilities, which are then combined according to user-defined wakeup word sequences, enabling customization without modifying the trained model.
2Adaptability or versatility
If a customized wakeup solution based on HMM decoding network is used, then the user customization requirement is met, but the computation complexity increases
Solution Approach 1:
The complex HMM decoding network is extracted and replaced with a simpler syllable combination sequence matching approach. Instead of using a full HMM decoding framework, the solution extracts only the necessary syllable recognition capability from the DNN model and combines syllables according to predefined sequences, significantly reducing computational complexity.
Solution Approach 2:
The syllable combination sequence is designed as a lightweight, easily replaceable structure that can be quickly constructed from the pronunciation dictionary without requiring complex model training. This disposable-like structure can be rapidly regenerated for different wakeup words, avoiding the high computational cost of retraining deep models.
3Adaptability or versatility
If a customized wakeup solution based on HMM decoding network is used, then the user customization requirement is met, but the response speed decreases
Solution Approach 1:
The syllable combination sequences are pre-constructed from the pronunciation dictionary before runtime detection. This preliminary preparation of valid syllable sequences enables fast matching during actual wakeup detection, as the system only needs to compare incoming speech syllables against the pre-combined sequences rather than performing complex decoding at runtime.
Data Source
AI summary
This application discloses an artificial intelligence-based (AI-based) wakeup word detection method performed by a computing device. The method includes: constructing, by using a preset pronunciation dictionary, at least one syllable combination sequence for self-defined wakeup word text inputted by a user; obtaining to-be-recognized speech data, and extracting speech features of speech frames in the speech data; inputting the speech features into a pre-constructed deep neural network (DNN) model, to output posterior probability vectors of the speech features corresponding to syllable identifiers; determine a target probability vector from the posterior probability vectors according to the syllable combination sequence; and calculate a confidence according to the target probability vector, and determine that the speech frames include the wakeup word text when the confidence is greater than or equal to a threshold.