Context-Dependent Phoneme Posterior Probability Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems face challenges in accurately classifying context-dependent phonemes, which affects overall classification accuracy.
Innovation Solution
A representation network with an input layer and multiple intermediate layers is trained to compute discriminative features, discretized, and hashed to generate posterior probabilities, improving phoneme classification accuracy by progressively computing features and assigning labels to nodes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional speech recognition systems use simple phoneme classification methods, then the system complexity is low, but the classification accuracy of context-dependent phonemes is insufficient
Solution Approach 1:
The patent segments the phoneme classification task into multiple intermediate layers, where each layer progressively refines the classification of context-dependent phonemes. This multi-layered approach divides the complex classification problem into manageable stages, improving accuracy while maintaining systematic complexity control.
Solution Approach 2:
The patent introduces an additional dimensional aspect by incorporating context information from surrounding phonemes and linguistic structures into the classification process. This multi-dimensional approach allows the system to capture contextual dependencies that single-dimension methods miss, thereby improving classification accuracy.
2Measurement precision
If the system uses multiple intermediate layers for progressive feature computation, then classification accuracy improves, but computational complexity increases
Solution Approach 1:
The computational process is segmented into multiple intermediate layers, each performing a specific feature computation task. This segmentation allows the system to progressively refine phoneme classifications through staged processing, improving accuracy while organizing computational complexity into manageable modules.
Solution Approach 2:
The system performs preliminary feature extraction and computation in earlier intermediate layers before final classification. By preparing and refining features in advance through multiple stages, the system reduces the computational burden on later stages and improves overall classification accuracy through progressive refinement.
Data Source
AI summary
What is disclosed is a system and method for modelling a class posterior probability of context dependent phonemes in a speech recognition system. A representation network is trained by projecting a N-dimensional feature vector into G intermediate layers of nodes. At least some features are associated with a class label vector. A last intermediate layer ZG of the representation network is discretized to obtain a discretized layer {circumflex over (Z)}. Feature vector Q is obtained by randomly selecting V features from discretized layer {circumflex over (Z)}. Q is repeatedly hashed to obtain a vector Qf where Qf is an output of the fth hashing. An equivalent scalar representation is determined for each Qf. In a manner more fully disclosed herein, a posterior probability Pf is determined for each (x, b) pair based on the equivalent scalar representation of each respective Qf. The obtained posterior probabilities are used to improve classification accuracy in a speech recognition system.