Gated Convolution Neural Network for Speech Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition methods lack an effective layer for extracting speech features, leading to relatively low accuracy in speech recognition due to inadequate feature extraction.
Innovation Solution
A method and device utilizing a gated convolution neural network to perform spectrum analysis and extract speech features, where the network consists of multiple convolutional layers with pooling layers, employing different non-linear activation functions to enhance feature extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a full connect layer is used to extract speech features in the speech recognition network, then the network structure is simple, but the speech recognition accuracy is low due to inadequate feature extraction
Solution Approach 1:
The patent segments the speech recognition network into distinct functional layers: a spectrum analysis layer for initial feature extraction, multiple convolutional layers for hierarchical feature learning, and a full connect layer for classification. This segmentation allows specialized feature extraction in the spectrum and convolutional layers, improving overall accuracy while maintaining reasonable complexity through modular design.
Solution Approach 2:
The patent transforms the speech signal from the time domain to the frequency domain through spectrum analysis before processing. This dimensional transformation enables the convolutional layers to operate on spectral features rather than raw waveforms, enhancing feature extraction capability and recognition accuracy without proportionally increasing complexity.
2Measurement precision
If multiple convolutional layers with different activation functions are added to improve feature extraction, then speech recognition accuracy improves, but the device complexity increases
Solution Approach 1:
The patent applies different non-linear activation functions (such as ReLU, Tanh, or Sigmoid) to different convolutional layers based on their specific functional requirements. This local differentiation optimizes feature extraction in each layer for its particular purpose while avoiding the need to redesign the entire network, thus improving feature extraction accuracy with controlled complexity increases.
Solution Approach 2:
The patent designs the network with continuous hierarchical feature extraction across multiple convolutional layers, where each layer builds upon the features extracted by the previous layer. This continuous processing ensures that useful feature extraction actions are maintained throughout the network depth, improving overall accuracy while the gradual feature transformation keeps complexity manageable.
Data Source
AI summary
Embodiments of the present disclosure provide a method and a device for extracting speech features based on artificial intelligence. The method includes performing a spectrum analysis on a speech to be recognized to obtain a spectrum program of the speech to be recognized; and extracting features of the spectrum program by using a gated convolution neural network to obtain the speech features of the speech to be recognized. As the spectrum program can describe the speech to be recognized in a form of image, and the gated convolution neural network is an effective method for processing images, the speech features extracted with this method may accurately describe characteristics of the speech.