Data processing: speech signal processing – linguistics – language – Speech signal processing – For storage or transmission
Reexamination Certificate
1999-04-19
2002-07-30
{haeck over (S)}mits, T{overscore (a)}livaldis Ivars (Department: 2741)
Data processing: speech signal processing, linguistics, language
Speech signal processing
For storage or transmission
C704S228000
Reexamination Certificate
active
06427133
ABSTRACT:
TECHNICAL AREA
This application is the national phase under 35 U.S.C. §371 of prior PCT International Application No. PCT/CH97/00289 which has an International filing date of Jul. 31, 1997 which designated the United States of America.
The invention concerns a procedure for evaluation of the transmission quality of a voice signal, with the voice signal being sent from a transmitter to a receiver by a digital transmission system with a specified data frame rate and the transmission being distorted, and with the transmitted and distorted voice signal being subjected to a subsequent evaluation.
STATE OF THE ART
It is known that the transmission quality of mobile radio systems can vary dramatically. Signal strengths at the receiving location, the geographical environment and under certain circumstances the travel speed of the mobile radio participant play a role in this. It is not clear how disruptions occurring on a physical level (radio transmission path) affect the digitally transmitted voice signal and its intelligibility. Nor is it entirely clear to which physical disruption a subjectively noticed impairment of the intelligibility or loss of quality of the voice signal reproduced by the receiver's end (through a loudspeaker) can be attributed.
From EP 0 644 674 A2, we know of a procedure for evaluating transmission quality of a voice transmission path that enables evaluation on an automatic level, which correlates heavily to human perception. I.e., the system can conduct an evaluation of the transmission quality and apply a scale as would be done by a trained test listener. Basically, an integral quality evaluation process takes place. The causes of quality losses are not investigated.
EP 0 722 164 A1 describes a system for signal quality evaluation where the signal (which is, for example, a voice signal scanned with 8 KHz) is divided into blocks or frames of a specified length N. The spectral output density is calculated for each frame. Further processing is based on the frequency sensitivity of the outer human hearing apparatus.
Of course the operator of a mobile radio system is not only interested in evaluating transmission quality, but also in improving it. It is therefore desirable to recognize certain acoustically noticeable effects as such and to evaluate them. For example, disruptions can occur that cause the voice signal to sound like the voice of a robot (“robot voice” effect). Often voice distortions occur as well, which lead to meowing fluctuations in the signal (“ping-pong” effect).
PRESENTATION OF THE INVENTION
It is an object of the invention to provide a procedure of the kind mentioned above, which enables the identification of certain acoustically noticeable distortions and their weighting with regard to a signal quality impairment.
According to the invention, frequency components, which correspond with a data frame rate of the digital transmission system, are extracted from the transmitted voice signal and analyzed. At a frame length of e.g. 20 ms, the lowest frequency that is analyzed is 50 Hz. Preferably, not only the 50 Hz component is extracted, but also its harmonics, i.e. its integral multiples.
Analysis and calculation of a ‘quality value’ can be conducted in various ways. For example, the individual amplitude values can be determined and compared to a reference value. Several amplitude values can be summarized. Also, spectral output density values can be calculated and compared to reference values. According to a preferred version, the reference value is determined on the basis of a previously conducted analysis of the signal that is to be transmitted. For this, the signal components according to the invention are filtered and analyzed as required in order to determine the reference value. Of course several reference values can be established, for example in order to be able to conduct a classification.
Preferably not all integral multiples of the base frequency are evaluated, but rather only a limited number of frequencies. The upper limit for frequencies that are to be analyzed is, for example, within the maximum sensitivity range of human hearing, i.e., at 2-3 KHz. This way, relevant acoustical effects can be identified and quantified without difficulty in most cases, while limiting the time spent on calculations. The base frequency itself does not need to be analyzed. The smallest harmonic to be considered can be at 200 Hz, for example. Generally all harmonics between the lowest and the highest frequency Hmin and Hmax are taken into consideration. It is also feasible, however, to evaluate only certain selected frequencies instead of an uninterrupted series of frequencies. The selection of these frequencies can result from an analysis of the communication system's transmission behavior and the respective spectral output.
It is best to process the voice signal through a sliding window, i.e., the signal's digital values are weighted according to a specified window function with the goal of keeping frequency components caused by the block formation as small as possible. From the scanned values gathered into a block, a weighted measurement value is determined. The length of such a window should preferably be 4 times larger than the data frame.
For the weighting process of the frequency components determined in accordance with the invention, frequencies of the transmitted signal that are in between can be used, i.e. for a grid of, for example, 50 Hz (50, 100, 150, 200, . . . Hz) preferably the frequencies in between these frequency intervals (75, 125, 175, . . . Hz) are filtered to determine the amplitude or output weight. This way, an output-independent factor for evaluating the received voice signal can be determined. In order to identify certain effects with characteristic frequency pictures, the extracted frequency components can also be weighted individually if necessary. The appropriate weighting function can be established in advance or calculated on a case-by-case basis.
With regard to the highest possible stability of the evaluation system it is beneficial if only the largest value of a series of reference values is used for the reception-related evaluation. In other words, when working with a sliding window and determining a reference value for each window position according to a specified calculation procedure, the value calculated in this way for a certain window position is not used immediately, but rather the neighboring values are taken into consideration as well in order to then take the largest of them, for example, as the effective reference value (local maximum).
The circuit design on the receiving end for execution of the procedure in accordance with the invention includes a processor to obtain discreet Fourier transforms and to analyze the previously established frequency components. The required reference values are memorized.
The outlet side of the described circuit design provides for a neural network. (This network may, but is not required to be, implemented in a separate circuit; i.e., it can also be in the processor itself or in software form.) The neural network is trained in advance with various sample signals so that it can conduct the desired evaluation in real time. The circuit design according to the invention can be applied as a supplement to the system described in EP 0 644 674 A2, i.e., the measurement values determined in accordance with the invention are entered into the neural network in addition to already known preliminary values.
The following detailed description and summary of patent claims shows further advantageous versions and combinations of features of the invention.
REFERENCES:
patent: 5715372 (1998-02-01), Meyers et al.
patent: 0644674 (1995-03-01), None
patent: 0722164 (1996-07-01), None
patent: 94 00922 (1994-01-01), None
patent: 95 15035 (1995-06-01), None
patent: 96 06496 (1996-02-01), None
Fahnle Thomas
Paping Martin
Armstrong Angela
Ascom Infrasys AG
{haeck over (S)}mits T{overscore (a)}livaldis Ivars
LandOfFree
Process and device for evaluating the quality of a... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Process and device for evaluating the quality of a..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Process and device for evaluating the quality of a... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2914455