Search

ClearSpeech: How to improve the accuracy of a voice recognition system by up to 29.7%?

ARKAMYS Dereverberation: How to improve the accuracy of a voice recognition system by up to 29.7%? Read to know more about this Dereverberation Module.

Posted 04/04/2016 on Sound Expertise

ARKAMYS ClearSpeech solution (Echo Cancellation and Noise Reduction), now integrates a best of breed feature developed by Dr. Nicolas Lopez, ARKAMYS ECNR Expert, called Dereverberation.

The Dereverberation software module uses one single microphone to accurately estimate and suppress reverberation, caused by the listening room (living room, car cabin, …), from voice recordings.

This module has been successfully showcased at the « Reverb Challenge » in Florence (IT) during the ICASSP Audio conference in May 2014.

For this benchmark, the Dereverberation module was used together with ARKAMYS Noise Reduction and it successfully improved the accuracy of a voice recognition system by up to 29.7% (expressed in Word Error Rate).

 

Find more details on this benchmark in our related publication [1].

[1] Lopez et al., “Reverberation Suppression Based on Sparse Linear Prediction in Noisy Environments”, REVERB Workshop 2014, Florence.

Want to know more about ARKAMYS ECNR solution: ClearSpeech

ARKAMYS – ECNR