Voice activity detector for an echo suppressor and an echo suppr

Telephonic communications – Echo cancellation or suppression

Patent

Rate now

  [ 0.00 ] – not rated yet Voters 0   Comments 0

Details

379389, 379390, 379409, 395 216, H04M 908, G10L 500

Patent

active

056195664

DESCRIPTION:

BRIEF SUMMARY
FIELD OF THE INVENTION

This invention relates to a voice activity detector suitable for use in an echo suppressor and, separately and in addition, it relates to an echo suppressor. The invention addresses the problem of the detection of voice (speech) in the presence of another speaker.
The need for voice activity detection arises, for example, in a (Vehicular) Speaker Phone (VSP) where the outbound (land to mobile) speech is introduced in the vehicle loudspeaker and the vehicle phone user may want to break in. Herein the abbreviation VAD is used for voice activity detection and voice activity detector, as the context requires.


BACKGROUND OF THE INVENTION

An arrangement which allows double talk in a full duplex communication channel is echo cancellation, where the outbound speech is cancelled at the inbound path. However, this arrangement is very expensive in terms of computation resources and thus is often not feasible.
At the other extreme, handling echoes in the vehicular environment is possible using the concept of an echo suppressor based on standard voice detection given a noisy background.
CCITT recommendation G.164 of 1988 Fascicle III.1 pages 186-205 describes generalised echo suppressors, among which a Type D echo suppressor is entirely digital and provides voice coding of a near-end voice and decoding (or synthesis) of a far-end voice. In such arrangements, "background speech" parameters are available if, for example, LPC type coding is performed as in many modern communication systems.
However, a conventional echo suppressor will not operate well in a full duplex situation where the loudspeaker power is comparable to the local user voice power, since its voice detection is based on the distinction between the speech power and/or its characteristics from those of the stationary background noise. Thus, either the loudspeaker voice will be detected by the VAD or the local user will be blocked when the far out VAD indicates activity.
Standard, present day VADs in the VSP context, are based on measuring the signal energy relative to background noise energy (Noise Riding Threshold (NRT)type detection). To implement such a detector, the VAD has to be able first to detect noise, estimate its instantaneous energy level, and subsequently detect speech if the signal energy exceeds a threshold (above the noise floor). In more advanced VADs the noise spectral characteristics are also estimated, and the energy out of the whitening inverse filter is used for the VAD. The process of distinguishing noise from voice is sometimes augmented with additional features extraction e.g. stationarity test and/or periodicity check (the noise being stationary and nonperiodic compared to speech).
GSM recommendation 06.32, 22 May 1989 is an example of energy based VAD where the energy is measured at the output of the inverse of the background noise shaping filter ("whitening" filter). The details, including the procedure for adapting the threshold and calculation of the filtered energy are given in that document.


SUMMARY OF THE INVENTION

According to a first aspect of the invention, a voice activity detector for an echo suppressor is provided, comprising: a receive audio path having a voice decoder for receiving voice parameters and synthesizing voice therefrom, a transmit audio path, a whitenrag filter for levelling the spectrum of the audio signal on the transmit path to provide a levelled signal, and decision means coupled to the whitening filter to measure energy in the levelled signal and thereby to detect voice on the transmit path, characterized in that the whitening filter is a voice whitening filter and means are provided for adapting the whitening filter according to the voice parameters received by the voice decoder in the receive path.
Thus the invention, in at least its first aspect, solves the problem of the detection of voice (speech) in the presence of another user by basing the detection on the other user speech characteristics.
In a second aspect of the invention, an echo suppressor is provided compr

REFERENCES:
patent: 3784747 (1974-01-01), Berkley et al.
patent: 4644108 (1987-02-01), Crouse et al.
patent: 4864608 (1989-09-01), Miyamoto et al.
patent: 5276765 (1994-01-01), Freeman et al.
patent: 5343521 (1994-08-01), Jullien et al.
patent: 5365583 (1994-11-01), Huang et al.
patent: 5475731 (1995-12-01), Rasmusson
Freeman et al., "The Voice Activity Detector for the Pan-European Digital Cellular Mobile Telephone Service", ICASSP 89, Glasgow, Scotland, 23 May 1989, vol. 1, pp. 369-372.
Jullien et al., "Acoustic Echo Controller for Wide-Band Hands-Free Telephony", Proc. of EUSIPCO-90, Sep. 18, 1990, Barcelona, Spain, pp. 1983-1986.
Wehrmann, "Concepts of Improving Hands-Free Speech Communication", IEEE International Symposium on Circuits and Systems, May 10, 1992, San Diego, CA, vol. 4, pp. 1918-1921.
Yasukawa et al., "An Acoustic Echo Canceller Using Subband Sampling and Decorrelation Methods", IEEE Transactions on Signal Processing, Feb. 1993, vol. 41, No. 2, pp. 926-930.

LandOfFree

Say what you really think

Search LandOfFree.com for the USA inventors and patents. Rate them and share your experience with other people.

Rating

Voice activity detector for an echo suppressor and an echo suppr does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Voice activity detector for an echo suppressor and an echo suppr, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Voice activity detector for an echo suppressor and an echo suppr will most certainly appreciate the feedback.

Rate now

     

Profile ID: LFUS-PAI-O-2402020

  Search
All data on this website is collected from public sources. Our data reflects the most accurate information available at the time of publication.