Voice Wake-Up Recognition Using Word and Syllable Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech interaction systems face challenges in achieving accurate and efficient wake-up recognition, particularly in terms of response speed, wake-up difficulty, and semantic understanding, which affect the smoothness of human-machine interaction.
Innovation Solution
A dual-stage recognition method involving word and syllable recognition is employed, using a convolutional neural network and a gate recurrent unit for word recognition, and a feature extraction and encoding module for syllable recognition, to enhance accuracy and reduce false positives, especially for short wake-up words.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only word recognition is used for wake-up detection, then the system response speed is fast, but the recognition accuracy is low and false positives occur frequently
Solution Approach 1:
The patent divides the wake-up recognition task into two independent stages: word recognition and syllable recognition. Each stage uses a separate neural network model (word recognition model and syllable recognition model) that processes the speech signal independently. This segmentation allows each model to be optimized for its specific task, improving overall accuracy while maintaining manageable complexity through modular design.
Solution Approach 2:
The patent performs both word recognition and syllable recognition, which is more than a single recognition system would provide. By implementing dual-stage recognition, the system performs partial recognition at the word level and additional verification at the syllable level, ensuring higher accuracy particularly for short wake-up words where single-stage recognition fails.
2Measurement precision
If dual-stage word and syllable recognition is implemented, then the recognition accuracy is improved, but the processing time increases
Solution Approach 1:
The system performs word recognition first as a preliminary filtering stage. The word recognition model quickly identifies potential wake-up words, and only when this stage indicates a positive match does the system proceed to the more time-consuming syllable recognition stage. This preliminary action reduces overall processing time by avoiding full dual-stage processing for non-wake-up speech.
Solution Approach 2:
By segmenting the recognition process into two independent models that can process speech frames in parallel, the system reduces processing time compared to sequential analysis. Each model processes speech features independently, allowing for efficient computation and faster overall response despite the dual-stage approach.
3Measurement precision
If traditional single-model recognition is used, then the device complexity is low, but the recognition fails for short wake-up words with few characters
Solution Approach 1:
The patent segments the recognition task into word-level and syllable-level analysis. The syllable recognition model specifically addresses short wake-up words by analyzing speech at the syllable granularity, which provides sufficient contextual information even for very short inputs. This segmentation enables accurate recognition of short words that would be indistinguishable from noise in single-model systems.
Solution Approach 2:
The syllable recognition model acts as an intermediary verification layer between the speech input and the final wake-up determination. For short wake-up words, the syllable-level analysis provides the necessary intermediate processing to distinguish valid wake-up commands from background noise, bridging the gap that single-model systems cannot cross.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
A speech wake-up method, an apparatus, an electronic device, a storage medium, and a program product are provided. The method includes: performing a word recognition on a speech to be recognized to obtain a wake-up word recognition result (S210); performing a syllable recognition on the speech to be recognized to obtain a wake-up syllable recognition result, in response to determining that the wake-up word recognition result represents that the speech to be recognized contains a predetermined wake-up word (S220); and determining that the speech to be recognized is a correct wake-up speech, in response to determining that the wake-up syllable recognition result represents that the speech to be recognized contains a predetermined syllable (S230).