Voice Bot Training Corpus Update for Intent Recognition

The patent replaces the mechanical intent schema matching system with a machine learning-based natural language understanding system. Instead of using predefined schemas to match user inputs, the system uses ML models to learn and adapt to various speech patterns and intents from training data, enabling the bot to handle unseen intents while maintaining reliable performance on defined actions.

The patent changes the fundamental parameter of how intent recognition is achieved - from using a large number of discrete intent schemas to using a machine learning model that processes speech continuously. This allows the system to capture speech nuances through learned patterns rather than explicit definitions, reducing the need for extensive computational resources while maintaining high versatility.

The patent substitutes the memory-intensive storage of large intent schema databases with a machine learning model that processes speech inputs dynamically. The ML model learns from training data and makes predictions without requiring storage of all possible intent schemas, significantly reducing memory requirements while maintaining the ability to handle diverse speech nuances.

The patent implements a self-service training approach where the machine learning model automatically learns from training data without requiring extensive manual definition and refinement of intent schemas. The system performs self-training by processing training instances and updating its internal models, achieving high accuracy while reducing the computational resources needed for manual configuration.