Skip to main content
European Commission logo print header

Speech Recognizer Quality Assessement for Linguistic Engineering

Exploitable results

In the SQALE project the applicability of various speech recognition systems to three European languages (British English, French and German) is assessed. Both the international speech recognition community, and the (recognition) assessment community may benefit from the results of this project. There exists a wide range of commercial speech recognition system providers who have focused on the American English language. The SQALE project may help them in making their systems work for other languages, and can serve as a standard for quality evaluation of their systems. The focus of the project is to compare 4 speech recognition systems in 4 languages. Good analysis methods are being developed for comparing systems and languages. Automatic dictation machines and voice response systems are the main applications that will benefit from the multilingual approach. A dictation machine is already in production. Of the 16 possible system-language combinations, 13 have been assessed, which is 5 more than planned. Word error rates for the strict conditions (limited training, vocabulary and language model) range from 13 % to 20 %. Statistical significance tests have been performed on the differences between systems and languages. A thorough statistical analysis is being carried out.

Searching for OpenAIRE data...

There was an error trying to search data from OpenAIRE

No results available