Customized Language Model for Speech Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition systems in contact centers face challenges in accurately recognizing phrases and speech patterns specific to certain industries due to variability in pronunciation, speed, pitch, and accent, leading to poor performance on context-specific words and phrases.
Innovation Solution
A method for customizing a speech recognition system by generating a customized language model using organization-specific training data, which includes both in-medium and out-of-medium data, to assign weights to generic training data based on similarities, thereby increasing the accuracy of phrase recognition and reducing errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a generic language model is used for speech recognition, then the system can handle a wide variety of speech inputs, but it fails to accurately recognize industry-specific phrases and terminology
Solution Approach 1:
The patent applies local quality by creating customized language models tailored to specific organizations and industries. Instead of using a single generic model, the system generates specialized models that incorporate organization-specific vocabulary, phrases, and speech patterns. This allows the speech recognition system to have high adaptability to industry-specific content while maintaining accurate recognition of contextual phrases through locally optimized language models.
2Measurement precision
If organization-specific training data is collected and processed, then speech recognition accuracy improves, but the system complexity and processing time increase
Solution Approach 1:
The patent implements preliminary action by pre-processing and organizing organization-specific training data before model generation. The system collects, cleans, and structures speech data in advance, creating ready-to-use training corpora that streamline the subsequent model customization process. This preliminary preparation reduces the complexity of real-time model generation and enables more efficient speech recognition accuracy improvement.
Solution Approach 2:
The patent uses an intermediary approach by introducing a language model generation module that acts as a mediator between raw organization-specific data and the speech recognition system. This intermediary component transforms unprocessed speech data into structured language models, simplifying the integration process and reducing overall system complexity while maintaining high recognition accuracy.
3Reliability
If multiple language models are generated for different organizations, then speech recognition becomes more accurate for each organization, but the computational resources and processing time required increase
Solution Approach 1:
The patent applies segmentation by dividing the language model generation process into modular components that can be independently processed. The system segments organization-specific data into distinct training corpora and generates specialized language models for each organization separately. This segmentation enables parallel processing of multiple models, reducing overall processing time while maintaining high recognition reliability for each individual organization.
Data Source
AI summary
A method for generating a language model for an organization includes: receiving, by a processor, organization-specific training data; receiving, by the processor, generic training data; computing, by the processor, a plurality of similarities between the generic training data and the organization-specific training data; assigning, by the processor, a plurality of weights to the generic training data in accordance with the computed similarities; combining, by the processor, the generic training data with the organization-specific training data in accordance with the weights to generate customized training data; training, by the processor, a customized language model using the customized training data; and outputting, by the processor, the customized language model, the customized language model being configured to compute the likelihood of phrases in a medium.