Voice Keyword Detection via Deep Learning Text Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice recognition technologies face challenges in achieving high recall rates and reducing false positives in vocal keyword retrieval, particularly in scenarios requiring sensitive voice information detection.
Innovation Solution
A deep learning-based voice data processing method that converts voice data into target text data using a preset voice recognition model, followed by keyword matching and homophone expansion, occurrence probability calculation, and text classification to determine the presence of target voice keywords with improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional voice recognition methods are used for keyword retrieval, then the system is simple and easy to implement, but the recall rate is low and false positives increase
Solution Approach 1:
The patent segments the keyword retrieval process into multiple stages: initial keyword matching, homophone expansion, occurrence probability calculation, and text classification. Each stage processes the voice data differently to progressively improve recall rate while managing system complexity through modular design
Solution Approach 2:
The patent performs preliminary homophone expansion and occurrence probability calculation before final keyword matching. By pre-processing the voice data to identify potential homophones and their probabilities, the system improves recall rate by capturing keywords that might otherwise be missed, while maintaining manageable complexity through staged processing
2Measurement precision
If traditional voice recognition methods are used, then the processing speed is fast, but the accuracy of keyword detection is low
Solution Approach 1:
The patent introduces an intermediary text classification model that acts as a mediator between voice data and keyword detection. This classifier processes the voice data through multiple layers (embedding, bi-directional LSTM, attention mechanism) to extract contextual features, thereby improving keyword detection accuracy while managing processing complexity through specialized intermediate processing
Solution Approach 2:
The patent transforms the one-dimensional keyword matching problem into a multi-dimensional analysis by incorporating: (1) temporal dimension through sequential processing of voice frames, (2) contextual dimension through bi-directional LSTM capturing past and future context, and (3) attention dimension through weighted focus on relevant time steps. This multi-dimensional approach improves detection accuracy while the modular architecture manages complexity
3Reliability
If multiple processing steps are added to improve recall rate, then keyword detection accuracy improves, but processing time increases
Solution Approach 1:
The patent implements periodic action through frame-based processing of voice data, where the system processes voice in discrete time frames with fixed window sizes. This periodic approach allows the system to apply multiple processing steps (homophone expansion, probability calculation, classification) to each frame independently, improving recall rate while managing processing time through parallelization across frames
Solution Approach 2:
The patent performs preliminary calculations of occurrence probabilities and homophone expansions before the main keyword matching process. By pre-computing these elements and storing them for quick reference during keyword detection, the system improves recall rate through comprehensive matching while reducing processing time by avoiding redundant calculations during the main detection phase
Data Source
AI summary
Disclosed of the present application is relation to deep learning based voice data processing. The voice data to be detected is converted into target text data based on a voice recognition model so that the keyword text corresponding to the predetermined target voice keyword can be converted. Then, the data is matched with the target text data to determine whether the voice data to be detected includes the target voice keyword based on the matching result. Thus, because the voice recognition model is obtained by deep learning based on the obtained voice recognition data training set, it can obtain high-precision target text data, thereby improving the accuracy of subsequent matching. The problem of low accuracy of detecting voice data for keyword detection can therefor be solved.