Multilingual Intent Recognition Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neural network-based speech-to-intent systems face challenges due to limited amounts of task-specific speech data, which exacerbates data sparsity issues, especially when automatic speech recognition requires transcription and intent labeling.
Innovation Solution
The approach involves multilingual training, where data from various languages is pooled to train a neural network model with shared layers common to all languages, and language-specific parameters are used to improve model training, allowing for better parameter estimation and intent classification across languages.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multilingual training data is used to train the neural network, then the model performance and parameter estimation improve, but the data processing complexity and training time increase
Solution Approach 1:
The patent segments the training data by language, organizing multilingual speech data into separate language-specific datasets. This segmentation allows the system to process and manage each language's data independently while still benefiting from multilingual training, thereby reducing the overall data processing complexity while maintaining model performance.
Solution Approach 2:
The patent implements a universal neural network model that can process multiple languages through shared layers and parameters. This multi-functionality allows the same model architecture to handle different languages without requiring separate models for each language, improving efficiency while maintaining high model performance across all languages.
2Measurement precision
If speech data is used for training instead of text data, then the intent recognition accuracy improves, but the data sparsity issue worsens due to limited task-specific speech data
Solution Approach 1:
The patent merges multilingual speech datasets into a unified training corpus, combining data from multiple languages to increase the overall quantity of available training data. This merging approach mitigates the data sparsity problem by leveraging the collective volume of multilingual speech data while maintaining language-specific characteristics through appropriate model architecture design.
Data Source
AI summary
Embodiments of the present invention provide computer implemented methods, computer program products and computer systems. For example, embodiments of the present invention can access one or more intents and associated entities from limited amount of speech to text training data in a single language. Embodiments of the present invention can locate speech to text training data in one or more other languages using the accessed one or more intents and associated entities to locate speech to text training data in the one or more other languages different than the single language. Embodiments of the present invention can then train a neural network based on the limited amount of speech to text training data in the single language and the located speech to text training data in the one or more other languages.