N-gram Language Model Tuning for Speech Recognition Precision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In automatic speech recognition systems, there is a trade-off between precision and recall in spoken term detection, where improving one parameter tends to decrease the other, making it challenging to achieve a balanced performance for specific customer demands or business operations.
Innovation Solution
A computer-implemented method for building an n-gram language model using training text data and additional text data with discount parameters, where the discount parameters are tuned using development data to achieve a predetermined balance between precision and recall, employing smoothing algorithms like Kneser-Ney Smoothing to redistribute n-gram counts and improve detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If confidence score threshold is increased to improve precision of spoken term detection, then precision is improved, but recall decreases
Solution Approach 1:
The patent applies parameter changes by adjusting the confidence score threshold and introducing discount parameters (D1, D2, D3+) in the smoothing algorithm to control the balance between precision and recall. By modifying these parameters, the system can optimize detection performance according to specific application requirements.
Solution Approach 2:
The patent implements dynamics by making the confidence threshold and discount parameters adjustable and tunable based on application-specific needs. The system allows dynamic optimization of precision-recall trade-off by training with development data and adjusting parameters accordingly, rather than using fixed thresholds.
2Measurement precision
If discount parameters are tuned to improve precision for target keywords, then precision is improved, but computational complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-training the language model with development data to optimize discount parameters before actual deployment. This pre-optimization process establishes the best parameter configuration in advance, reducing the need for complex real-time computations during actual spoken term detection.
Solution Approach 2:
The system performs self-service by automatically tuning discount parameters through training with development data. The optimization process is automated, allowing the system to self-adjust parameters for optimal performance without requiring manual intervention or complex external optimization tools.
Data Source
AI summary
A method, a system, and a computer program product for building an n-gram language model for an automatic speech recognition. The method includes reading training text data and additional text data both for the n-gram language model from a storage, and building the n-gram language model by a smoothing algorithm having discount parameters for n-gram counts. The additional text data includes plural sentences having at least one target keyword. Each discount parameter for each target keyword is tuned using development data which are different from the additional text data so that a predetermined balance between precision and recall is achieved.