# US 5,708,754 A · Method for real-time reduction of voice telecommunications noise not measurable at its source

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 5,708,754 A
- Document ID: US5708754A
- Status: Granted
- Kind: A
- Application number: 08/790,006
- Filed: 1997-01-28
- Granted: 1998-01-13
- Earliest claimed: 1993-11-30
- Assignees: AT&T Corp.
- Inventors: Woodson Dale Wynn
- Web page: https://patentgrep.com/patent/US5708754A

## CPC

- G10L 21/0208: Noise filtering
- H04M 3/18: with means for reducing interference {or noise}; with means for reducing effects due to line faults {with means for protecting lines}
- H04M 3/40: Applications of speech amplifiers
- G10L 21/0264: characterised by the type of parameter measurement, e.g. correlation techniques, zero crossing techniques or predictive techniques
- G10L 25/12: the extracted parameters being prediction coefficients
- G10L 25/24: the extracted parameters being the cepstrum
- G10L 2021/02168: {the estimation exclusively taking place during speech pauses}
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- H04M: TELEPHONIC COMMUNICATION (circuits for controlling other apparatus via a telephone cable and not involving telephone switching apparatus G08)

## Abstract

A telecommunications network service overcomes the annoying effects of transmitted noise by a signal processing which filters out the noise using interactive estimations of a linear predictive coating speech model. The speech model filter uses an accurate updated estimate of the current noise power spectral density, based upon incoming signal frame samples which are determined by a voice activity detector to be noise-only frames. A novel method of calculating the incoming signal using the linear predictive coating model provides for making intraframe iterations of the present frame based upon a selected number of recent past frames and up to two future frames. The processing is effective notwithstanding that the noise signal is not ascertainable from its source.

## Claims

### Claim 1

In a telecommunications network carrying an incoming signal, said incoming signal having both speech and noise energy, a method for real-time processing of said incoming signal using an iterated filter operated as a Linear Predictive Coding speech model comprising filter-order adjustments and means for equivalently representing pole positions of said Linear Predictive Coding speech model with Line Spectral Pair position roots and difference roots, in which said iterated filter creates an estimate of the speech power spectrum component of said incoming signal for each current frame for transmission to said telecommunications network, said method comprising the steps of:
converting said incoming signal to a time-series of spectral domain data frames;
selecting from said time-series of data frames a subset of consecutive data frames including a group of past data frames, the current data frame and up to two future data frames;
performing a selected number of iterations upon said current data frame across said subset of data frames, each said iteration generating values of said Line Spectral Pair position roots and difference roots;
during said iteration sequence, for each iteration smoothing said Line Spectral Pair root positions generated for said past data frames, with the Line Spectral Pair root position of said current frame, using values of Line Spectral Pair root positions calculated for the current iteration of said sequence;
creating estimates of said incoming speech power spectrum from the output of a succession of said iteration sequences of said current frames;
forming a noise-reduced speech signal based on said incoming speech power spectrum estimates; and
transmitting said noise-reduced speech signal to said telecommunications network.

### Claim 2 (depends on 1)

The process of claim 1, wherein said iterations comprise the steps of:
creating a succession of Wiener Filters in which the signal spectrum entering the initial said Wiener Filter includes a noise power spectrum estimate; and
at each present iteration stage, determining the best current estimates of speech power spectrum and noise power spectrum, and summing the two said estimates to form a current said Wiener Filter.

### Claim 3 (depends on 2)

The process of claim 2, wherein the number of said past frames included in each said iteration is five or less.

### Claim 4 (depends on 3)

The process of claim 3, wherein the number of iterations performed on each said current frame is not greater than 5.