Adaptive Proper Name Recognition in Speech Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automatic speech recognition (ASR) and natural language understanding (NLU) systems struggle with accurately recognizing and understanding proper names, especially uncommon ones, leading to incorrect results in spoken commands.
Innovation Solution
The system employs a primary recognizer for initial transcription, followed by a natural language understanding module to identify potential proper names, which are then processed by a secondary recognizer adapted using an adaptation object generator to recognize novel names and entities, allowing for accurate transcription and understanding of proper names with low latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a primary recognizer is used for initial transcription, then the system can process general speech, but it fails to accurately recognize uncommon proper names
Solution Approach 1:
The system divides the recognition process into two segments: a primary recognizer for general speech transcription and a secondary recognizer specialized for proper names. The NLU module segments the transcription to identify potential proper names, which are then passed to the secondary recognizer for accurate recognition. This segmentation allows each component to specialize in its strength.
Solution Approach 2:
The NLU module acts as an intermediary between the primary recognizer and the secondary recognizer. It processes the initial transcription, identifies potential proper names, and prepares them for the secondary recognizer. This intermediary step enables the system to leverage both general speech recognition capabilities and specialized proper name recognition.
2Measurement precision
If the ASR system is extensively adapted to recognize proper names, then recognition accuracy improves, but computational costs and economic expenses increase
Solution Approach 1:
Instead of adapting the entire ASR system, the patent applies specialized adaptation only to the secondary recognizer that handles proper names. This local quality approach allows the system to maintain general speech recognition capabilities while adding targeted proper name recognition without the full computational burden of system-wide adaptation.
Solution Approach 2:
The system performs partial adaptation by creating a secondary recognizer with adapted grammars and vocabularies specifically for proper names, rather than fully adapting the entire ASR system. This partial action achieves the necessary proper name recognition accuracy without the excessive computational costs of complete system adaptation.
3Measurement precision
If a secondary recognizer with adaptation is introduced, then proper name recognition accuracy improves, but processing time increases
Solution Approach 1:
The NLU module performs preliminary action by identifying and segmenting potential proper names from the initial transcription before they are passed to the secondary recognizer. This preliminary processing allows the secondary recognizer to focus only on relevant segments, reducing overall processing time while maintaining high accuracy.
Solution Approach 2:
The transcription process is segmented so that only portions containing potential proper names are processed by the computationally intensive secondary recognizer. This segmentation significantly reduces the amount of data requiring specialized processing, thereby reducing latency while maintaining recognition accuracy.
Data Source
AI summary
Various embodiments contemplate systems and methods for performing automatic speech recognition (ASR) and natural language understanding (NLU) that enable high accuracy recognition and understanding of freely spoken utterances which may contain proper names and similar entities. The proper name entities may contain or be comprised wholly of words that are not present in the vocabularies of these systems as normally constituted. Recognition of the other words in the utterances in question—e.g., words that are not part of the proper name entities—may occur at regular, high recognition accuracy. Various embodiments provide as output not only accurately transcribed running text of the complete utterance, but also a symbolic representation of the meaning of the input, including appropriate symbolic representations of proper name entities, adequate to allow a computer system to respond appropriately to the spoken request without further analysis of the user's input.