Semi-Word Acoustic Model for Speech Recognition Vocabulary Contraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large language vocabularies in speech recognition systems require significant storage space and result in limited accuracy due to their unbounded nature and continuous addition of new words, making it challenging to efficiently process and recognize human speech.
Innovation Solution
The system restructures conventional speech recognizer data by segmenting pronunciation dictionary entries into constituent phonemes and grouping them based on initial phonemes, creating a semi-word acoustic model that reduces the number of possibilities and improves recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a large unbounded vocabulary is used to accommodate continuously added new words, then the system can recognize more words, but storage space requirements increase and recognition accuracy decreases
Solution Approach 1:
The patent segments words into phoneme sequences and groups them by common prefixes (initial N phonemes). This segmentation allows the system to handle large vocabularies by breaking them into manageable chunks that can be stored more efficiently and processed in a structured manner, reducing the storage burden while maintaining vocabulary coverage.
Solution Approach 2:
The patent changes the parameter of vocabulary organization from storing complete words to storing phoneme sequences with prefix grouping. By transforming the representation parameters (from whole words to segmented phoneme structures), the system achieves better storage efficiency while preserving the ability to recognize a large number of words.
2Adaptability or versatility
If a large unbounded vocabulary is used to accommodate continuously added new words, then the system can recognize more words, but recognition accuracy decreases
Solution Approach 1:
By segmenting words into phoneme sequences and grouping by common prefixes, the system creates a more structured representation that improves acoustic modeling. This segmentation allows for better statistical modeling of phoneme patterns, which enhances recognition accuracy even as vocabulary size grows.
Solution Approach 2:
The transformation from whole-word storage to phoneme-sequence storage with prefix grouping changes the fundamental parameters of vocabulary representation. This parameter change enables more effective acoustic-phonetic modeling, which directly improves recognition accuracy by capturing phonetic patterns more effectively.
3Adaptability or versatility
If conventional pronunciation dictionaries are used for large vocabularies, then all words can be represented, but processing efficiency decreases
Solution Approach 1:
The segmentation of vocabulary into phoneme sequences with prefix grouping enables more efficient processing by reducing the search space. When processing speech, the system can quickly identify potential matches by matching phoneme sequences and prefixes, rather than searching through entire word forms, significantly improving processing efficiency.
Solution Approach 2:
By changing the organizational parameters from complete words to segmented phoneme structures, the system optimizes processing efficiency. The transformed representation allows for faster acoustic matching and more efficient decoding, as the structured phoneme sequences enable better utilization of acoustic models and faster candidate generation.
Data Source
AI summary
A speech recognition system includes, or has access to, conventional speech recognizer data, including a conventional acoustic model and pronunciation dictionary. The speech recognition system generates restructured speech recognizer data from the conventional speech recognizer data. When used at runtime by a speech recognizer module, the restructured speech recognizer data produces more accurate and efficient results than those produced using the conventional speech recognizer data. The restructuring involves segmenting entries of the conventional pronunciation dictionary and acoustic model according to their constituent phonemes and grouping those entries with the same initial N phonemes, for some integer N (e.g., N=3), and deriving a restructured dictionary with a corresponding semi-word acoustic model for the various grouped entries. The decomposition of the conventional pronunciation dictionary into the restructured dictionary with semi-word acoustic model greatly reduces the number of possibilities in the dictionaries (e.g., from potentially unlimited to finite and relatively small), and also improves the accuracy of speech recognition.