Entity Name Recognition via Phonetic Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional speech recognizers often inaccurately transcribe proper names in voice commands, leading to erroneous outputs, as seen in the example where 'Navigate to CITYZEN restaurant in Washington DC' is transcribed as 'Navigate to CITIZEN restaurant in Washington DC', resulting in nonsensical results.
Innovation Solution
An enhanced natural language processing system that receives an utterance, identifies a proper name by matching phonetic representations with pre-associated canonical names from an entity directory, using a knowledge base to expand entity names and generate phonetic representations, and verifies the transcription using an entity type-specific, geo-localized database to ensure accurate recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional speech recognizers are used to transcribe voice commands, then the system is simple and fast, but the transcription accuracy of proper names deteriorates
Solution Approach 1:
The patent introduces phonetic representations as an intermediary layer between the speech recognizer and the entity database. The system converts transcribed names into phonetic form, queries entities with matching phonetic representations, and retrieves canonical names. This intermediary approach allows accurate entity recognition without requiring the speech recognizer itself to be highly accurate at proper names.
Solution Approach 2:
The system pre-generates phonetic representations for all canonical entity names and stores them in an entity database before processing user queries. This preliminary preparation enables rapid matching during runtime without performing complex phonetic analysis on-the-fly, resolving the contradiction between accuracy and processing speed.
2Measurement precision
If phonetic matching with canonical names is implemented, then proper name recognition accuracy improves, but processing time increases
Solution Approach 1:
Phonetic representations of all canonical entity names are pre-computed and stored in the entity database. When a user query arrives, the system only needs to convert the transcribed name to phonetic form and perform a database lookup, rather than performing complex phonetic analysis during query processing. This eliminates most of the processing time overhead.
Solution Approach 2:
The system creates phonetic copies of canonical entity names and stores them alongside the original names in the entity database. During query processing, the system compares the phonetic copy of the transcribed name against stored phonetic representations, enabling fast matching without repeatedly performing phonetic analysis on canonical names.
3Measurement precision
If the system uses an entity directory with canonical names, then transcription accuracy improves, but the complexity of data management increases
Solution Approach 1:
The patent introduces phonetic representations as an intermediary that bridges the gap between user speech and canonical entity names. Users speak names in any form, the system converts them to phonetic form, matches against pre-stored phonetic representations of canonical names, and returns the correct canonical form. This intermediary layer shields the system from the complexity of managing multiple name variations.
Solution Approach 2:
The system transforms entity names from their original textual form into phonetic representations for storage and comparison purposes. By changing the parameter space from text matching to phonetic matching, the system can handle variations in spelling, capitalization, and pronunciation without requiring complex text normalization rules.
Data Source
Figure 1
Figure 2
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for recognizing names of entities in speech. In one aspect, a method includes actions of receiving an utterance that includes (i) a first term that indicates a particular entity type, and (ii) a second term that indicates an entity name. Additional actions include obtaining a phonetic representation of the second term and determining that the phonetic representation of the second term matches a particular phonetic representation of a particular canonical name of a set of canonical names associated with a particular entity. Further actions include outputting a reference name associated with the particular entity as a transcription of the second term.