Spelling Recognition Using Speech and Keypad Input
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automatic speech recognition systems face challenges in accurately recognizing spellings and long digit-strings over the telephone, especially when the target population's name distribution does not match the training data, leading to high out-of-vocabulary rates and error rates.
Innovation Solution
The system combines speech input with non-speech input from a telephone keypad or other modalities, dynamically constructing a weighted grammar using statistical letter models to improve recognition accuracy, allowing for disambiguation and reducing error rates by leveraging keypad sequences and directory lookups.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If domain constraints such as database lookups are used to improve recognition accuracy, then spelling recognition accuracy is improved, but the system cannot handle out-of-vocabulary cases and reduces adaptability
Solution Approach 1:
The system dynamically adjusts the recognition approach based on whether the input matches known vocabulary. For in-vocabulary items, it uses constrained database lookup for high accuracy. For out-of-vocabulary items, it switches to unconstrained statistical recognition methods, allowing the system to adapt its behavior to the specific input context and maintain both accuracy and versatility.
2Ease of operation
If speech input alone is used, then the system is simple and easy to operate, but spelling recognition accuracy deteriorates due to inherent difficulty in recognizing long digit-strings and spellings
Solution Approach 1:
The system merges speech recognition with keypad input recognition to create a hybrid system. The speech module handles natural language input while the keypad module provides precise character confirmation. The combination leverages the naturalness of speech with the precision of keypad input, achieving high accuracy without sacrificing ease of operation since users can switch between modalities as needed.
3Measurement precision
If keypad input is used for disambiguation, then spelling recognition accuracy is improved, but the complexity of the system increases due to integrating multiple input modalities
Solution Approach 1:
The system introduces a grammar construction intermediary that translates keypad input sequences into constrained grammars for speech recognition. This intermediary layer reconciles the two different input modalities by converting keypad sequences into speech recognition constraints, allowing the system to leverage both input types without direct complex integration. The intermediary simplifies the overall system architecture while maintaining high accuracy.
4Adaptability or versatility
If unconstrained speech recognition is used, then the system is adaptable to any vocabulary, but recognition accuracy deteriorates for difficult vocabularies such as letters and digits
Solution Approach 1:
The system dynamically adjusts the recognition approach based on whether the input matches known vocabulary. For in-vocabulary items, it uses constrained database lookup for high accuracy. For out-of-vocabulary items, it switches to unconstrained statistical recognition methods, allowing the system to adapt its behavior to the specific input context and maintain both accuracy and versatility.
Data Source
AI summary
A system and method for non-speech input or keypad-aided word and spelling recognition is disclosed. The method comprises performing spelling recognition via automatic speech recognition (ASR) on received speech from a user, the ASR being performed using a statistical letter model trained on domain data and producing a letter lattice RLN. If an ASR confidence is below a predetermined level, then the method comprises receiving non-speech input from the user, generating a keypad constraint grammar K and generating a letter string based on a composition of finite state transducers RLN and K. Other variations of the invention include recognizing input by first receiving non-speech input, dynamically generating an unweighted grammar, generating a weighted grammar using domain data, and then performing speech, and thus spelling, recognition on input speech using the weighted grammar.