SLU Model Training Without Full Transcripts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current spoken language understanding (SLU) systems require word-for-word transcripts for training, which are time-consuming and computationally expensive, and are not efficient in processing speech data from diverse domains like customer calls without specific order of entities.
Innovation Solution
The system trains SLU models using natural language training data that includes speech recordings and semantic entities/intents without transcripts, employing attention-based encoder-decoder neural networks and connectionist temporal classification (CTC) models, allowing for efficient computational processing and reduced data volume.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If word-for-word transcripts are used for training SLU systems, then training data completeness is improved, but computational cost and time consumption increase
Solution Approach 1:
The patent extracts only the essential semantic entities and intents from speech data, removing the need for complete word-for-word transcripts. This extraction approach retains the critical information needed for SLU training while eliminating redundant transcription data, thereby reducing computational cost and time consumption.
Solution Approach 2:
The patent uses lightweight semantic annotations (entities and intents) instead of expensive, time-consuming full transcripts. These simplified training data representations are easier and faster to generate, providing a cost-effective alternative that maintains training effectiveness without requiring complete linguistic transcription.
2Loss of information
If word-for-word transcripts are used for training SLU systems, then training data completeness is improved, but computational complexity increases
Solution Approach 1:
The patent extracts only the essential semantic entities and intents from speech data, removing the need for complete word-for-word transcripts. This extraction approach retains the critical information needed for SLU training while eliminating redundant transcription data, thereby reducing computational cost and time consumption.
Solution Approach 2:
The patent uses lightweight semantic annotations (entities and intents) instead of expensive, time-consuming full transcripts. These simplified training data representations are easier and faster to generate, providing a cost-effective alternative that maintains training effectiveness without requiring complete linguistic transcription.
3Adaptability or versatility
If semantic entities are processed without specific order, then adaptability to diverse domains is improved, but difficulty of processing increases
Solution Approach 1:
The patent changes the parameter of entity ordering from fixed (spoken order) to flexible (any order including alphabetical). This parameter change allows the system to adapt to diverse domains where entities may appear in different orders, while the underlying processing methodology remains consistent and manageable.
Data Source
AI summary
A method and system of training a spoken language understanding (SLU) model includes receiving natural language training data comprising (i) one or more speech recording, and (ii) a set of semantic entities and/or intents for each corresponding speech recording. For each speech recording, one or more entity labels and corresponding values, and one or more intent labels are extracted from the corresponding semantic entities and/or overall intent. A spoken language understanding (SLU) model is trained based upon the one or more entity labels and corresponding values, and one or more intent labels of the corresponding speech recordings without a need for a transcript of the corresponding speech recording.