Cross-Language G2P Conversion for Voice Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition systems face challenges in handling multiple pronunciations for words across different languages, particularly for native speakers of Mandarin and Cantonese, and in contexts where words from different languages are combined, requiring a labor-intensive process to generate a Grammar and Dictionary (GnD) that accounts for pronunciation differences.
Innovation Solution
A method and system for cross-language Grapheme to Phoneme (G2P) conversion that automatically generates an optimized set of pronunciations for a target language, using an improved data structure and process to handle multiple languages, including pronunciation refining and phoneme transformation rules to consolidate pronunciations, allowing for efficient handling of hybrid language recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a Grammar and Dictionary is manually generated to account for pronunciation differences across languages, then recognition accuracy for multiple languages is improved, but the process becomes labor-intensive and time-consuming
Solution Approach 1:
The system pre-generates multiple pronunciation variants for each word in the dictionary during an offline preparation phase. These pre-computed pronunciations cover different language accents and variations, so that during actual speech recognition, the system can directly match against these pre-prepared options without requiring manual intervention for each new language scenario.
Solution Approach 2:
The system automatically generates and updates pronunciation dictionaries by detecting pronunciation patterns from training data and self-adapting to new languages and accents. The speech recognition system itself performs the work of creating language-specific pronunciation rules through machine learning, eliminating the need for manual linguist intervention.
2Adaptability or versatility
If multiple pronunciations are associated with each word in the Grammar and Dictionary, then cross-language recognition capability is improved, but the complexity of the data structure increases
Solution Approach 1:
The pronunciation dictionary is segmented into language-specific subsets, each containing pronunciations tailored to particular languages or accents. Instead of one monolithic complex structure, the system divides the dictionary into manageable language-specific modules that can be independently processed and matched, reducing the complexity burden on any single data structure.
Solution Approach 2:
The system introduces an intermediary layer that maps between phonetic representations and language-specific pronunciations. This intermediary structure acts as a bridge, allowing the core dictionary to remain relatively simple while handling complex cross-language variations through the intermediary mapping layer that translates between different phonetic systems.
Data Source
Figure 1
Figure 2A~2B
Figure 3
AI summary
Computer implemented speech processing generates one or more pronunciations of an input word in a first language by a non-native speaker of the first language who is a native speaker of a second language. The input word is converted into one or more pronunciations. Each pronunciation includes one or more phonemes selected from a set of phonemes associated with the second language. Each pronunciation is associated with the input word in an entry in a computer database. Each pronunciation in the database is associated with information identifying a pronunciation language and/or a phoneme language.