Grammar-Based Speech Recognizer for Alternate Pronunciation Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems are limited in handling alternate pronunciations and mispronunciations, particularly for non-native speakers, as they rely on general pronunciation dictionaries that do not cover all possible pronunciations, leading to recognition errors.
Innovation Solution
A system and method that utilizes a grammar-based recognizer, acoustic models, and phoneme confusion matrices to detect and learn alternate pronunciations by training on a large speech corpus, constructing phoneme replacement candidate lists, and integrating these with linguistic dictionaries to create an improved pronunciation dictionary, while also detecting accents and mispronunciations at phone and syllable levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a general pronunciation dictionary is used for speech recognition, then the system is simple to operate, but it cannot handle alternate pronunciations and mispronunciations, leading to recognition errors
Solution Approach 1:
The system performs preliminary actions by training acoustic models on large speech corpora before actual speech recognition. The grammar-based recognizer pre-processes speech data to identify potential mispronunciations and alternate pronunciations, then learns from these examples to improve recognition accuracy for non-native speakers and accented speech.
Solution Approach 2:
The system implements feedback mechanisms where the grammar-based recognizer continuously learns from speech data and updates the pronunciation dictionary. Mispronunciations detected during speech recognition are fed back into the system to train acoustic models, which then provide improved pronunciation information in subsequent recognition tasks, creating a continuous improvement loop.
2Reliability
If the pronunciation dictionary is expanded to include alternate pronunciations, then recognition accuracy improves, but the dictionary size and processing complexity increase
Solution Approach 1:
The system applies local quality by making the pronunciation dictionary dynamic and adaptive rather than static. The grammar-based recognizer identifies and learns alternate pronunciations specific to individual words and contexts, updating the dictionary locally rather than expanding it uniformly. This allows the system to handle non-standard pronunciations where needed while maintaining a manageable dictionary size overall.
Solution Approach 2:
The pronunciation dictionary is made dynamic through continuous learning from speech data. The system automatically updates the dictionary with new alternate pronunciations and mispronunciations identified during speech recognition tasks, allowing the dictionary to adapt to different speakers, accents, and contexts without manual intervention or fixed expansion.
3Measurement precision
If phoneme-level mispronunciation detection is implemented, then recognition precision improves, but the processing time and computational requirements increase
Solution Approach 1:
The system segments the speech recognition task into multiple stages: initial speech recognition, phoneme-level analysis for mispronunciation detection, and pronunciation learning. The grammar-based recognizer handles phoneme-level detection separately from word-level recognition, allowing precise mispronunciation identification without requiring complete reprocessing of the entire speech signal, thus managing computational requirements.
Data Source
AI summary
A system and method for learning alternate pronunciations for speech recognition is disclosed. Alternative name pronunciations may be covered, through pronunciation learning, that have not been previously covered in a general pronunciation dictionary. In an embodiment, the detection of phone-level and syllable-level mispronunciations in words and sentences may be based on acoustic models trained by Hidden Markov Models. Mispronunciations may be detected by comparing the likelihood of the potential state of the targeting pronunciation unit with a pre-determined threshold through a series of tests. It is also within the scope of an embodiment to detect accents.