Data processing: speech signal processing – linguistics – language – Speech signal processing – Psychoacoustic
Reexamination Certificate
2001-03-09
2004-04-06
Dorvil, Richemond (Department: 2654)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Psychoacoustic
C704S230000, C455S423000, C455S067120, C375S224000
Reexamination Certificate
active
06718296
ABSTRACT:
This invention relates to the measurement of quality of a sound signal, and more specifically a speech signal. Objective processes for this purpose are currently under development and are of application in prototype testing, pre-delivery testing of components, and in-service testing of installed equipment. They are most commonly used in telephony, but are also of application in other systems used for carrying speech signals, for example public-address systems.
The present applicant has a number of patents and applications relating to this technical field, most particularly European Patent 0647375, granted on Oct. 14
th
1998. In this system, a signal degraded by the system under test is compared with a reference signal, which has not passed through the system under test, to identify audible errors in the degraded signal. These audible errors are assessed to determine their perceptual significance—that is, errors of types which are considered significant by human listeners are given greater weight than are those which are not considered so significant. Since only audible errors are assessed, inaudible errors, which are perceptually irrelevant, are not assessed.
The automated system provides an output comparable to subjective quality measures originally devised for use by human subjects. More specifically, it generates two values, Y
LE
and Y
LQ
, equivalent to the “Mean Opinion Scores” (MOS) for “listening effort” and “listening quality”, which would be given by a panel of human listeners when listening to the same signal, as will be discussed later. The use of an automated system allows for more consistent assessment than human assessors could achieve, and also allows the use of compressed and simplified test sequences, and multilingual test sequences, which give spurious results when used with human assessors because such sequences do not convey intelligible content.
Such automated systems require a known (reference) signal to be played through a distorting system (the telephone network) to derive a degraded signal, which is compared with an undistorted version of the reference signal. Such systems are known as “intrusive” measurement systems, because whilst the test is carried out the system under test cannot carry live (revenue-earning) traffic.
An auditory transform of each signal is taken, to emulate the response of the human auditory system (ear and brain) to sound. The degraded signal is then compared with the reference signal in the perceptual domain, in which the subjective quality that would be perceived by a listener using the network is determined from parameters extracted from the transforms.
A suitable test signal is disclosed in International Patent Specification WO/95/01011 (EP0705501) and comprises a sequence of speech-like sounds, selected to be representative of the different types of phonetic sounds that the system under test may have to handle, presented in a sequence. The sounds are selected such that typical transitions between individual phonetic elements are represented. Typical speech comprises a sequence of utterances separated by silent periods, as the speaker pauses to breathe, or listens to the other party to the conversation. These silent periods, and the transitions between utterances and silent periods, are also modelled by the test signal.
This existing system reliably assesses most speech carrier technologies employed within conventional analogue and digital switched telephone networks. In such networks a dedicated connection is provided between the two parties to a call, for the duration of that call, and all speech is carried over that connection. However, connectionless packet-based speech transmission systems are beginning to be introduced, in particular for use in the “Internet” and companies' internal “Intranets”. In a connectionless packet-based system each transmission is divided into a series of data “packets”, which travel independently from one user to the other. Intermediate nodes in the network transmit the packets to each other according to address information carried in each packet. However, according to the demands of other traffic on the various links between such nodes, and the available capacity on those links, different packets may be delayed, or may travel by different routes to reach the same destination. Consequently, end-to-end times vary from one packet to another. For the transmission of data such as text, or downloading of computer files for the recipient to use subsequently, such variations in end-to-end times are of little consequence. However, when used for real-time speech, these variations can affect the clarity of the speech as perceived by the user.
Various proposals have been made to try to minimise the delay to a level which does not interfere with conversation and comprehension—see for example the present applicant's International Patent Application WO99/12329, and the article by R Barnett in “
Electronics and Communication Engineering Journal
”, October 1997, entitled “
Connectionless ATM
”. However, it is fundamental to such connectionless systems that some variation in the residual delay will occur. A single speech utterance is typically assembled from the information carried in several packets. However, variations in the delay between individual packets will in general not be apparent in the resulting utterance, as the slowest packet generally determines the delay to the utterance as a whole. However, the delay to each complete utterance can vary considerably between one utterance and the next, as buffer lengths are normally adjusted during periods of silence.
Changes to the delay occurring during the course of an utterance, for example because part of the utterance is missing, will be more apparent in the resulting utterance.
In addition to changes in residual delay, transmission systems are now beginning to come into use in which changes in other characteristics, such as level (signal amplitude), can occur. See ITU-T draft recommendation G. 169.
The human brain is insensitive to small changes in delay and amplitude between speech events, so these variations may be imperceptible to a human listener, provided the magnitude of the effect is not such as to interfere with conversation. However, the prior art measuring system is sensitive to such variations, so that it returns unreliable values for signal quality when testing connectionless packet systems—that is, the results do not accurately reflect the subjective quality reported by human subjects.
If the delay is constant, the two signals can easily be synchronised to take account of the delay. However, if the degraded signal suffers variable delay, at least some parts of the degraded signal would not be synchronised with the test signal. The lack of synchronisation in those parts would be detected as substantial errors, which would be so great as to mask any errors caused by actual degradation of the signal. This would lead to an inaccurate measure of the subjective effect of the degradation.
There is therefore a requirement for a measurement system that is robust against such variable delays.
According to the invention, there is provided apparatus for testing equipment for handling speech signals, comprising
means for receiving first and second signals, means for selecting individual sections in the first signal and second signal,
means for comparing each section in the second signal with the corresponding section in the first signal to generate a distortion perception measure which indicates the extent to which the distortion of said section would be perceptible to a human listener, and
means for combining the results of each such measurement to generate an overall measure of the extent to which the distortion of the second signal with respect to the first signal would be perceptible to a human listener.
Preferably, the overall measure takes account of the perceptual importance of each section. The perceptual importance of a given section will depend on the number of individual speech components, and their relative importance to subjectiv
Beamond Eleanor J
Hollier Michael P
Philip Gray
Reynolds Richard J B
Rix Antony W
British Telecommunications public limited company
Dorvil Richemond
Harper V. Paul
Nixon & Vanderhye P.C.
LandOfFree
Measurement of signal quality does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Measurement of signal quality, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Measurement of signal quality will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3208271