Generation and Pruning of Pronunciation Variants to Improve ASR Accuracy
This work addresses a specific challenge in automatic speech recognition for phone services, offering an incremental improvement over existing methods.
The paper tackled the problem of pronunciation variations in speech recognition, particularly for names, by proposing a data-driven technique that automatically learns and updates pronunciation dictionaries, resulting in a 42% error rate reduction on a database with over 13,000 human names.
Speech recognition, especially name recognition, is widely used in phone services such as company directory dialers, stock quote providers or location finders. It is usually challenging due to pronunciation variations. This paper proposes an efficient and robust data-driven technique which automatically learns acceptable word pronunciations and updates the pronunciation dictionary to build a better lexicon without affecting recognition of other words similar to the target word. It generalizes well on datasets with various sizes, and reduces the error rate on a database with 13000+ human names by 42%, compared to a baseline with regular dictionaries already covering canonical pronunciations of 97%+ words in names, plus a well-trained spelling-to-pronunciation (STP) engine.