EP0722164A1
Expired / Lapsed
EP
⏱️ 0.0 years remaining
Method and apparatus for characterizing an input signal
Assignee: AT&T Inc
•
Filing: 1995-12-20
•
Granted: —
•
Expires: 2015-12-20
Abstract
Abstract The present invention provides a method and apparatus for measuring at least one signal characteristic. Initially, a set of features is selected which characterize a signal. An intelligent system, such as a neural network, is trained in the relationship between feature sets and signal characteristics. The selected feature set is then extracted from a first input signal. The extracted feature set from the first signal is input to the trained intelligent system. The intelligent system creates an output signal based on the feature set extracted from the first input signal. This output signal is then used to characterize the input signal. In one embodiment, the invention assesses voice quality, typically as expressed in MOS scores, in a manner which accurately corresponds to the analysis of human evaluators. For voice signals processed by voice coders, the present invention provides a measurement technique which is independent of various voice coding algorithms and consistent for any given algorithm.
Inventors
- 👤 Martin H. Meyers
- 👤 Carl F. Weaver
- 👤 Ahmed A. Tarraf
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-11-08 18:24
CPC Classifications
G; G01; G01R; G01R29/00; G01R29/26; G10; G10L; G10L15/00; G10L15/02; H; H04; H04M; H04M3/00; H04M3/22; G10L25/00; G10L25/27; G10L25/30; H04M2201/00; H04M2201/40
Claims Analysis
Total Claims: 21
Claim 1
Independent Claim
83 words
A method for measuring a characteristic of a signal comprising: selecting a set of features which characterize a signal; extracting the feature set from a first input signal; inputting the feature set extracted from the first input signal to an intelligent system which has been trained in the relationship between feature sets and signal characteristics, the intelligent system creating an output signal based on the feature set extracted from the first input signal; and using the output signal to characterize the input signal.
Claim 2
Dependent Claim
34 words
A method in accordance with claim 1 wherein selecting said set of features includes choosing one or more metrics representative of the signal characteristic and calculating the feature set based upon the chosen metrics.
Claim 3
Dependent Claim
37 words
A method in accordance with claim 2, wherein utilizing the output signal further includes averaging a weighted sum of a plurality of the output signals from the intelligent system to provide an average measure of signal characteristics.
Claim 4
Dependent Claim
15 words
A method in accordance with claim 1 wherein the intelligent system comprises a neural network.
Claim 5
Dependent Claim
19 words
A method in accordance with claim 1 wherein the intelligent system is trained by use of a backpropagation algorithm.
Claim 6
Dependent Claim
19 words
A method in accordance with claim 1 wherein the intelligent system is trained by use of a momentum algorithm.
Claim 7
Dependent Claim
21 words
A method in accordance with claim 1 wherein the intelligent system is trained by use of an adaptive learning rate algorithm.
Claim 8
Dependent Claim
15 words
A method in accordance with claim 1 wherein the input signal comprises a voice signal.
Claim 9
Dependent Claim
16 words
A method in accordance with claim 1 wherein the input signal comprises a coded voice signal.
Claim 10
Independent Claim
115 words
A method for measuring the voice quality of a voice signal by means of a neural network, said neural network having an input port and producing at least one output signal, the method comprising: providing a voice signal; selecting one or more metrics representative of voice signal quality; creating a feature set based on said selected metrics; extracting said feature set from said voice signal, training a neural network in the relationship between feature sets and a corresponding measure of voice signal quality; presenting the feature set extracted from said voice signal to an input port of said neural network; and using the neural network to create an output signal indicative of voice signal quality.
Claim 11
Dependent Claim
18 words
A method in accordance with claim 10, wherein said neural network training includes use of a backpropagation algorithm.
Claim 12
Dependent Claim
19 words
A method in accordance with claim 10, wherein said neural network training includes the use of a momentum algorithm.
Claim 13
Dependent Claim
21 words
A method in accordance with Claim 10, wherein said neural network training further includes use of an adaptive learning rate algorithm.
Claim 14
Dependent Claim
31 words
A method in accordance with claim 10 wherein the output signal is averaged using a weighted sum with a plurality of output signals to provide a measure of voice signal quality.
Claim 15
Independent Claim
66 words
A method for measuring signal quality comprising; selecting a feature set based on selective metrics representative of a signal; training a neural network in the relationship between the feature set and corresponding subjectively-obtained criteria; extracting the feature set from an input signal; inputting the extracted feature set to the neural network; and utilizing an output of the neural network to produce a measure of signal quality.
Claim 16
Dependent Claim
18 words
A method in accordance with claim 12 wherein the selected metrics are based on a power spectral measure.
Claim 17
Dependent Claim
16 words
A method in accordance with claim 15 wherein the subjectively-obtained criteria is based on MOS scoring.
Claim 18
Dependent Claim
59 words
A method in accordance with claim 15 wherein the feature extraction includes obtaining a coded and uncoded version of the signal, frame blocking both the coded and uncoded versions of the signal, determining the power spectral densities of the frame blocked signal, filtering the determined power spectral densities, and establishing the signal-to-noise ratio between the filtered power spectral densities.
Claim 19
Dependent Claim
89 words
Apparatus for measuring a characteristic of a signal comprising: means for selecting a set of features which characterize a signal; means for extracting the feature set from a first input signal; input means for inputting the feature set extracted from the first input signal to an intelligent system which has been trained in the relationship between feature sets and signal characteristics; and means for creating an output signal based on the feature set extracted form the first input signal, the output signal comprising information for characterizing the input signal.
Claim 20
Dependent Claim
14 words
Apparatus in accordance with claim 19 wherein the intelligent system comprises a neural network.
Claim 21
Dependent Claim
27 words
Apparatus in accordance with claim 20 wherein the neural network comprises a multilayer perceptron including an input layer, one or more hidden layers, and an output layer.
References Cited (Backward Citations)
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.