Monthly Archives: October 2013

New Feature – Speech Accuracy

A big challenge when learning a new language is knowing whether or not what you are saying could actually be understood by a native speaker.

If you attend a language class, your teacher can quickly give you feedback, but you might have to fight with other students for their time and perhaps you might be a little shy or afraid to make mistakes. If you are studying on your own, then you are likely in a worse situation being not able to even get this minimal feedback.

At OpenLanguage we take pride in our engaging lessons, which provide lesson input, and our software tools, that help provide review opportunities, but we also realize facilitating practice opportunities and providing corrective feedback is critical for developing your new language skills.

To help with this, today we are adding a new ‘Accuracy’ feature to the Sentence and Word Review tools available in our iOS and Android mobile apps.


To take advantage of this feature, simply tap the Accuracy button to get started. To hear a native speaker read the term or sentence press the play button. Then it is your turn to try. Press record and read the term yourself. The app will analyze your recording and give you an Accuracy score.


Now, this feature utilizes speech-recognition technology which is notoriously not 100% perfect. You should approach your score, not as an absolute measure of your performance, but more as an indicator of how likely you would be understood by a native speaker. The higher your score, the more likely you will be able to engage in a conversation with that native speaker.

This feature is simply designed to be a low-risk way for you to get feedback on how well you are pronouncing those foreign-language terms.

From a technical perspective, the app converts the spoken audio to a text string (with Android using the Android SDK and with iOS using Nuance) and then compares the resultant text string with the original for the native speaker. The app then uses a string comparison algorithm (Levenshtein distance) to determine an approximate accuracy score.

If you have any suggestions please let us know in the comments.