ASR Language Model for Foreign Word Transliteration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional language models for automatic speech recognition (ASR) and text-to-speech (TTS) systems face challenges in recognizing and pronouncing foreign words and proper nouns, leading to inappropriate responses due to database quality and algorithm prediction issues, particularly with Korean users encountering Latin-based or Chinese characters.
Innovation Solution
An electronic device generates an ASR language model that includes information about multiple candidate transliterations for variously utterable text, based on user context, basic language models, and customized language models, updating the model in response to user utterances to provide personalized voice recognition and voice output.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a conventional language model with rules and dictionaries is used, then the system structure is simple, but the recognition accuracy of foreign words and proper nouns deteriorates
Solution Approach 1:
The patent introduces a specialized language model module that acts as an intermediary between the conventional ASR/TTS systems and foreign words or proper nouns. This module contains dedicated databases and algorithms for handling transliterations and various pronunciations of foreign characters, intercepting and processing these specific cases before they reach the conventional recognition systems, thereby improving accuracy without requiring complete system redesign
Solution Approach 2:
The patent segments the language processing system into distinct functional modules: a conventional ASR/TTS system for general processing, and a specialized foreign word processing module for handling Latin-based words, Chinese characters, and proper nouns. This segmentation allows each module to be optimized independently, maintaining system simplicity while enhancing specific functionality
2Reliability
If a comprehensive database of various pronunciations is built, then the recognition performance improves, but the data collection effort and cost increase
Solution Approach 1:
The patent creates a universal language model module that handles multiple types of foreign words and proper nouns (Latin-based words, Chinese characters, Korean characters, Japanese characters) using a unified framework. This multi-functional module processes various character types and pronunciation variations through common algorithms and data structures, reducing the need for separate specialized databases for each language or character type
Solution Approach 2:
The system performs preliminary processing of foreign words and proper nouns by pre-building a language model with common transliteration patterns and pronunciation variations. This preliminary action prepares the system in advance to handle these cases efficiently during actual speech recognition, reducing the need for extensive real-time data collection and processing
3Measurement precision
If the language model is customized for each user, then the voice recognition accuracy improves, but the processing time increases
Solution Approach 1:
The patent applies local quality by customizing the language model specifically for foreign words and proper nouns in user contact information, while leaving the processing of common words handled by the conventional fast ASR system. This selective customization ensures that time-consuming personalized processing is applied only where necessary (foreign words and proper nouns), maintaining overall processing efficiency while improving accuracy for critical cases
Data Source
AI summary
An example electronic device may include a memory configured to store instructions and a processor electrically connected to the memory and configured to execute the instructions. When the instructions are executed by the processor, the processor may be configured to create an automatic speech recognition (ASR) language model including information about a plurality of candidate transliterations for a variously utterable text, based on a context of a user indicating a situation of the user, a basic language model, or a customized language model and update the customized language model in response to an utterance of the user matching one of the plurality of candidate transliterations.