eu | es | fr | en
Elhuyar Speech recognition system

ASR or Automatic Speech Recognition technology consists of converting speech into text. In this page you will find information on the speech recognition service for Basque made available by the language technologies department of Elhuyar. We can integrate this service into web services or applications, adapted to each's specificities.

This service is based on the AhoSR system developed by the Aholab group of the UPV/EHU.

This service is in use in the following websites:

Contact: Igor Leturia Azkarate (


It works on any computer (Linux, Windows, Mac), in the most recent versions of Firefox, Chrome, Opera and Edge. It does not work in Internet Explorer and Safari.

It works on Android, in the most recent versions of Firefox, Chrome and Opera.

It does not work on iOS, in any browser.

Instructions for use

When clicked, the speech recognition system asks for permission to accesss the microphone. We must allow access to it, otherwise it wil not work. Some browsers give the option not to ask everytime, by giving permission for always or remembering last choice.

After clicking on the recognition button, we have to wait until the button becomes red before speaking. Speak as soon as it becomes red, without waiting.

In computers, we need a microphone or a computer with a microphone jack. The better the microphone, the better the system will work. It works best with the microphones that come with headphones and that stand near the mouth. The volume of the microphone should be properly adjusted, not too high and not too low.

In mobile phones, do not put the telephone near your mouth, just keep it at normal reading distance.

If there is too much noise or interferences, the system does not work well. You do not have to speak loudly (on the contrary, if you speak too loud it gets saturated and works worse), just speak normal.

Speak clearly. While the recognition is going on, do not repeat the word(s), wait until it recognizes what has been said, and if it does not recognize anything then try again.

Like in any dictionary, do not add the article ("-a") at the end of the word.