Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Progressive Variable Window for Noise-Floor Determination

IP.com Disclosure Number: IPCOM000106816D
Original Publication Date: 1993-Dec-01
Included in the Prior Art Database: 2005-Mar-21
Document File: 4 page(s) / 144K

Publishing Venue

IBM

Related People

Hsieh, D: AUTHOR [+3]

Abstract

Described is a method for determining the level of background noise in an active information carrying communications channel. The method implements an improved real-time algorithm for detecting the presence of speech and/or estimating the level of noise floor. Two aspects are presented: Sampling windows of increasing width; and peak detection using a statistical approach.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 42% of the total text.

Progressive Variable Window for Noise-Floor Determination

      Described is a method for determining the level of background
noise in an active information carrying communications channel.  The
method implements an improved real-time algorithm for detecting the
presence of speech and/or estimating the level of noise floor.  Two
aspects are presented:  Sampling windows of increasing width; and
peak detection using a statistical approach.

Applications, such as speaker phones and telephone answering
machines, often make use of a voice detector in its functions.  In
speakerphones, it is necessary to monitor both the local phone set
and the phone line to determine the listener and talker functions.
One method uses fixed thresholds to determine the active function.
The thresholds would be adjusted such that speech over a certain
preset level on either the phone set, or the phone line would cause
that end of the channel to be the transmitting side, provided that
the other side was not active.  While a fixed threshold approach is
considered adequate, there are limitations for some applications,
such as the inability to compensate for signal levels, or the
background noise level on the phone line, or even the acoustic noise
level of the rooms in which the conversation is originating.  As a
result, variations can degrade the performance of the speakerphone.

      The above limitations of the threshold approach can be overcome
by allowing the thresholds to vary and to adapt to the conditions
present on the communications channel.  One method is to approximate
the background noise level on the channel, which is the sum of the
electrical noise on the channel and the acoustic noise picked up by
the originating microphone.  If the signal on the channel exceeds the
background noise by a certain margin, then the presence of speech can
be assumed.  Unfortunately, since the background noise and the voice
signal occupy the same frequency band, it is difficult to separate
them when they are present simultaneously.  However, by recognizing
that voice is discontinuous, in that the speech contains periods of
silence, there is an opportunity to measure the noise level of the
channel by quantifying the signal level, or noise, during pauses in
speech.

      In prior art, an algorithm was used for recognizing and
measuring the channel noise [*].  This algorithm attempted to measure
the noise of the channel and then it multiplies the estimated noise
by a fixed factor so as to yield a threshold value.  If the signal
exceeds the threshold during a period of time, speech is said to be
present.  In this algorithm, digital samples of analog speech are
taken and during a short period of time - approximately sixteen
milliseconds, the magnitude of each sample is calculated and the
largest magnitude sample during this period is stored as the peak
magnitude sample.  This process is repeated continuously.

      A second process then examines the p...