Context-Aware Speech Recognition Training to Reduce Hallucinations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional speech recognition systems generate text content that is not sufficiently accurate due to insufficient integration of context information.
Innovation Solution
A method and apparatus for context-based speech processing that involves obtaining training data including speech samples, context information, and annotation text, determining output probabilities using a speech recognition model with and without context information, and adjusting model parameters based on the difference between these probabilities to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If context information is integrated into speech recognition, then accuracy is improved, but model complexity increases
Solution Approach 1:
The patent segments the speech recognition process into two distinct processing paths: one that processes speech features with context information and another that processes speech features without context information. This segmentation allows the model to handle complex context-integrated recognition while maintaining a simpler baseline path, thereby improving accuracy without proportionally increasing overall model complexity.
Solution Approach 2:
The patent introduces an intermediary mechanism by computing the difference between outputs from the two processing paths (with and without context). This difference representation serves as a mediator that captures the contextual contribution separately, allowing the system to leverage context information for improved accuracy while keeping the additional computational burden localized and manageable.
2Manufacturing precision
If context information is integrated into speech recognition, then precision of generated text is improved, but computational resources increase
Solution Approach 1:
The patent implements partial action by selectively applying context information processing only to the extent necessary for improving precision. The dual-path architecture allows the system to process speech through both context-dependent and context-independent paths, using only the additional computational resources needed for the context-enhanced path, rather than applying context processing uniformly to all recognition tasks.
Solution Approach 2:
The patent changes the parameter of context integration by computing the difference between context-integrated and non-integrated outputs. This parameter transformation allows the system to capture contextual benefits while representing the additional computational requirement as a controlled difference term, enabling efficient resource utilization while maintaining improved text precision.
Data Source
AI summary
Embodiments in the disclosure relate to context-based speech processing. In an example method provided by the disclosure, training data is obtained, including a speech sample, context information associated with the speech sample, and annotation text corresponding to the speech sample. A first output probability corresponding to the annotation text is determined by processing a first feature sequence using a speech recognition model. The first feature sequence is constructed based on the speech sample and the context information. A second output probability corresponding to the annotation text is determined by processing a second feature sequence using the speech recognition model. The second feature sequence is constructed based on the speech sample and is independent of the context information. A training loss based on at least a difference between the first output probability and the second output probability is determined to adjust a parameter of the speech recognition model.