Semantic Tagger for Speech Recognition Word Lattices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems lack the ability to accurately tag and interpret semantic information in spoken words, leading to reduced accuracy and limited functionality in automated transcription and command recognition.
Innovation Solution
A method is introduced to tag words in a word lattice using a semantic tagger, which involves distilling a large named entity disambiguation model into a lightweight tagger, constructing joint and conditional distributions of n-grams, and applying these to generate transcriptions with semantic tags, enabling better accuracy and additional downstream functionality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large named entity disambiguation model is used for semantic tagging, then tagging accuracy is improved, but processing speed and computational efficiency deteriorate
Solution Approach 1:
The patent extracts and applies only the necessary semantic tagging components from a large named entity disambiguation model, rather than using the entire model. This selective extraction maintains tagging accuracy for relevant entities while reducing computational overhead and improving processing speed.
Solution Approach 2:
The patent changes key parameters of the large model to optimize performance, including adjusting confidence thresholds, modifying tagging granularity, and tuning probability cutoffs. These parameter adjustments allow the system to maintain high tagging accuracy while reducing computational complexity and improving processing speed.
2Measurement precision
If semantic tags are added to word lattice, then transcription accuracy is improved, but system complexity increases
Solution Approach 1:
The patent segments the speech recognition system into distinct modules: acoustic modeling, language modeling, and semantic tagging. Each module operates independently with well-defined interfaces, allowing semantic tags to be added without fundamentally redesigning the entire system architecture. This modular segmentation manages complexity while enabling accuracy improvements.
Solution Approach 2:
The patent introduces semantic tags as an intermediary layer between word recognition and transcription generation. These tags provide additional contextual information without requiring direct modification of core recognition algorithms, thereby improving transcription accuracy while adding minimal system complexity through a non-intrusive intermediate component.
3Reliability
If conditional probability transducer is constructed from supervised training corpus, then tagging reliability is improved, but training data requirements and processing time increase
Solution Approach 1:
The patent performs preliminary action by pre-training the conditional probability transducer on a supervised training corpus during an offline phase. This pre-computation of probabilities and tagging rules allows the system to achieve high tagging reliability during deployment without requiring extensive real-time processing, thereby reducing operational training time while maintaining reliability.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for tagging during speech recognition. A word lattice that indicates probabilities for sequences of words in an utterance is obtained. A conditional probability transducer that indicates a frequency that sequences of both the words and semantic tags for the words appear is obtained. The word lattice and the conditional probability transducer are composed to construct a word lattice that indicates probabilities for sequences of both the words in the utterance and the semantic tags for the words. The word lattice that indicates probabilities for sequences of both the words in the utterance and the semantic tags for the words is used to generate a transcription that includes the words in the utterance and the semantic tags for the words.