Method and apparatus for locating a talker

Abstract

An apparatus for locating a talker, comprising a microphone array for receiving multiple audio signals, wherein the microphone array is characterized by a predetermined beampattern, a spectral conditioner for filtering the audio signals to optimize the beampattern for talker localization, a localization estimator for calculating a localization estimate based on the filtered audio signals, an activity detector for detecting periods of speech activity, and decision logic for verifying the localization estimate during the periods of speech activity.

Description

CROSS REFERENCE TO RELATED APPLICATIONS [0001] This application is a divisional of U.S. patent application Ser. No. 09 / 894,539, filed on Jun. 28, 2001, now pending, which claims priority under 35 U.S.C. §119 to United Kingdom Patent No. 0016142.2, filed on Jun. 30, 2000, the contents of which are herein incorporated by reference in their entirety for all purposes.FIELD OF THE INVENTION [0002] The present invention relates generally to audio systems and in particular to a method and apparatus for locating a talker in a noisy or reverberant environment. BACKGROUND OF THE INVENTION [0003] Localization of sources is required in many applications, such as teleconferencing, where the source position is used to steer a high quality microphone beam toward the talker. In video conferencing systems, the source position may additionally be used to focus a camera on the talker. [0004] It is known in the art to use electronically steerable arrays of sensors in combination with location estimator...

AI Technical Summary

Benefits of technology

[0006] According to the present invention, spectral conditioning is used to enhance the performance of the estimator algorithm by restricting operation of the estimator to a narrow frequency band chosen to optimize localization rather than acoustic audibility, in contrast with the prior art. An activity detector is also used, as known from the prior art, to identify voiced segments. However, according to an important aspect of the invention, a decision logic state machine is implemented for receiving information from the activity detector and spectrally conditioned estimator and in response verifying localization estimates during periods of voice activation.

[0007] The method and apparatus of the present invention result in much faster talker localization than prior art approaches (typically less than 40 milliseconds), and require much less computational power. The use of spectral conditioning results in increased resolution (i.e. how close two talkers are allowed to be while ensuring accurate localization). Furthermore, the method and apparatus of the present invention are characterized by high operational stability in the presence of noise.

Problems solved by technology

Localization of acoustic sources is fraught with practical difficulties.

Firstly, reflecting walls (or other objects) generate virtual acoustic images of the source, which can be misidentified as real sources by the location estimator algorithm.

Secondly, most of the known locator estimator algorithms are unable to distinguish between noise sources and talkers, especially in the presence of correlated noise and during speech pauses.

Images