Multi-Microphone Speech Enhancement

Abstract

Speech signals recorded in a reverberant environment are often distorted by reverberation and ambient noise. Researchers at the International Audio Laboratories Erlangen have developed a large family of speech enhancement technologies suitable for a variety of applications including audio conferencing systems, speaker phones, and smart TVs. On this page some of our multi-microphone speech enhancement technologies are demonstrated.

For more information about these technologies please contact Prof. Dr. Emanuël Habets (emanuel.habets@audiolabs-erlangen.de).

Note: If you experience any problems viewing this page, please try to use the latest version of Firefox.

Audio-Video Examples

Joint Directional Filtering and Dereverberation

Measurements were carried out in an office room with reverberation time approximately 300 ms using a non-uniform linear array with four omnidirectional microphones. The distance between the array and the speaker is approximately 2.5-3 m. Only the array geometry and the location of the desired speaker was assumed to be known. The aim is to extract the desired speaker while reducing noise and reverberation.


Dereverberation

Measurements were carried out in a large conference room with reverberation time approximately 600 ms using a non-uniform linear array with four omnidirectional microphones. The total width of the microphone array was 12.25 cm. Two male speakers were moving in the room at a distance of 2-6 m from the array. No prior information, such as source positions or room characteristics, was assumed. The aim is to reduce both ambient noise and reverberation without distorting the speech.

Dereverberation: Scenario 1


Dereverberation: Scenario 2


Dereverberation: Scenario 3

Audio Examples

Measurements were carried out in a large conference room with reverberation time approximately 600 ms using a non-uniform linear array with four omnidirectional microphones. The total width of the microphone array was 12.25 cm. Two male speakers were moving in the room at a distance of 2-6 m from the array. No prior information, such as source positions or room characteristics, was assumed. The aim is to reduce both ambient noise and reverberation without distorting the speech.

The audio player allows you to easily switch between the unprocessed and processed signal.

Scenario 1

Scenario 2

Scenario 3