Speech Recognition of Prepositive Words Using Dual Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition technologies struggle to accurately recognize short words containing prepositive words due to their limited decoding search space and the lack of specific optimization in large-scale continuous speech recognition frameworks, leading to high manpower and time costs for re-training models.
Innovation Solution
A speech recognition method that extracts a first feature containing prepositive word information and a two-dimensional time-frequency domain feature from a sample speech signal, using a trained naive Bayes classifier to determine the presence of prepositive words without requiring retraining of the acoustic and language models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing speech recognition technology is used for short words containing prepositive words, then the decoding search space is limited, but the recognition accuracy deteriorates
Solution Approach 1:
The patent segments the speech recognition process into two distinct stages: a fast stage using a limited decoding search space for initial processing, and a slow stage using a complete decoding search space for final accuracy. This segmentation allows the system to benefit from both the efficiency of limited search and the accuracy of complete search, resolving the contradiction between decoding search space and recognition accuracy for short words containing prepositive words.
2Adaptability or versatility
If large-scale continuous speech recognition frameworks are used, then the model coverage is comprehensive, but the re-training cost increases
Solution Approach 1:
The patent performs preliminary action by pre-processing speech signals to identify and extract short words containing prepositive words before the main recognition process. This preliminary identification allows the system to apply specialized processing only where needed, rather than re-training entire large-scale models, thus maintaining model coverage while significantly reducing re-training time and computational costs.
3Measurement precision
If traditional speech recognition methods are used for short words, then the processing speed is fast, but the recognition accuracy deteriorates
Solution Approach 1:
The patent implements a dynamic two-stage recognition system where the processing approach adapts based on the characteristics of the input speech. For short words containing prepositive words, the system dynamically switches to a more thorough recognition process in the slow stage, while maintaining fast processing for other cases. This dynamic adaptation resolves the contradiction between processing speed and recognition accuracy.
Data Source
AI summary
Provided are a speech recognition method, a speech recognition apparatus, a computer readable storage medium, and an electronic device. The method comprises: obtaining a sample speech signal, decoding the sample speech signal, obtaining a decoding result, and extracting a first feature from the decoding result; extracting a target speech segment from the sample speech signal, obtaining a log magnitude spectrum of the target speech segment, and determining a second feature according to the log magnitude spectrum; combining the first feature and the second feature to obtain a third feature; training an untrained classifier by using the third feature so as to obtain a trained classifier; and obtaining a third feature to be recognized of a speech signal to be recognized, so as to determine whether the third feature to be recognized comprises a prepositive word.