Skip to content

News Arts and Science Teaching Media Library Services IEM - intern Contact
  You are not logged in Link icon Log in
You are here: Home » Kunst & Forschung » Akustik » Aufnahme und Wiedergabetechnik » Implementierung eines Telefonkonferenzsystems in PD basierend auf einer koinzidenten Mikrofonanordnung

Implementierung eines Telefonkonferenzsystems in PD basierend auf einer koinzidenten Mikrofonanordnung

Karl Freiberger

A common way to improve sound quality in teleconferencing systems is to pick up sound preferentially in the direction of the current speaker. Sound from other directions, on the other hand, shall be suppressed. This requires two tasks to be accomplished: 1) Acoustic Source Localization (ASL), i.e. estimating the position of the sound source 2) Beamforming, i.e. steering the microphone focus in a given direction Usually, microphone arrays are used for that purpose. This work investigates a communication front-end that incorporates a planar 4-channel coincident microphone array. Compared with usual arrays, the coincident array is very compact and handy. Two loudspeakers are used to play back the communication partner. The goal is to achieve the following demands at the same time: 1) Flawless pick up of the speaker 2) Reduction of diffuse noise, i.e. maximizing the directivity index 3) Suppression of the loudspeakers, i.e. null steering in the direction of the loudspeakers Coincident steering is reviewed for general setups, which for instance includes the soundfield ambisonic microphone. Furthermore, the independent component analysis (ICA) is addressed and its results are interpreted in terms of a directivity pattern. In addition to a full-duplex null-steering scheme, also a half-duplex echo suppression method is considered. Here, the microphone is turned off as soon as the loudspeakers are detected as the sound-source. The half-duplex mode is only reasonable under the following, realistic conditions: 1) The LS-positions are known and differ significantly from the speaker position. 2) No simultaneous talking Both are advantageous in full-duplex mode as well, because the ASL-algorithm does not explicitly account for multiple sources. Besides simulations in Matlab, the system is implemented in the real-time audio signal processing environment Pure Data (PD). As a starting point, ASL-routines for Matlab are available. Thus, the particular focus of this project lies in beamforming and implementation of the overall system.

Karl Freiberger    type: TI-Project    state: finished project     Date: 11.10.2009

Last modified 15.01.2010