Trained artificial neural networks using an imperfect vocal trac

Details Trained artificial neural networks using an imperfect vocal trac Trained artificial neural networks using an imperfect vocal trac

1998-02-03

2000-03-07

Hudspeth, David R.

Data processing: speech signal processing, linguistics, language

Speech signal processing

For storage or transmission

704228, 706 20, H04M 322, H04B 346, G01L 918

Patent

active

060352701

DESCRIPTION:

BRIEF SUMMARY
BACKGROUND OF THE INVENTION

1. Field of the Invention
This invention relates to the assessment of an audio signal carrying speech. It is of particular application to the assessment of the condition of telecommunications systems whilst in use.
2. Related Art
Signals carried over telecommunications links can undergo considerable transformations. such as digitisation, data compression, data reduction, amplification, and so on. All of these processes can distort the signals. For example. in digitising a waveform whose amplitude is greater than the maximum digitisation value, the peaks of the waveform will be converted to a flat-topped form (a process known as peak clipping). This adds unwanted harmonics to the signal. Distortions can also be caused by electromagnetic interference from external sources.
The distortions introduced by the processes described above are non-linear, so that a simple test signal may not be distorted in the same way as a complex waveform such as speech, or at all. For a telecommunications link carrying data it is possible to test the link using all possible data characters (e.g. the two characters 1 and 0 for a binary link, or the twelve tone-pairs used in DTMF (dual tone multi-frequency) systems. However speech does not consist of a limited number of well-defined signal elements, but is a continuously varying signal, whose elements vary according to not only the content of the speech (and the language used) but also the physiological and psychological characteristics of the individual speaker, which affect characteristics such as pitch, volume, characteristic vowel sounds etc.
It is known to test telecommunications equipment by running test sequences using samples of speech. Comparison between the test sequence as modified by the equipment under test and the original test sequence can be used to identify distortion introduced by the equipment under test. For example, Edmund Quincy, in the IEEE International Conference on Communications 87; Session 33.3; vol 2 (pages 1164-1171) describes such a method of analysing such a signalm using a "rule-based" system (also known as an "expert" system), in which predetermined objective rules are used to generate, for a given input signal, an appropriate output indicative of the quality of the signal.
The arrangement described above requires the use of a pre-arranged test sequence, which means it cannot be used on a live telecommunications link--that is, a link currently in use for revenue-earning traffic--because the test sequence would interfere with the traffic being carried and be audible to the users, and because conversely the live traffic itself (whose content cannot be predetermined) would be detected by the test equipment as distortion of the test signal.
In order to carry out tests on equipment in use, without interfering with the signals being carried by the equipment (so-called non-intrusive testing), it is desirable to carry out the tests using the live speech signals themselves as the test signals. However, a problem with using live speech as the zest signal is that there is no instantaneous way of obtaining, at the point of measurement, a sample of the original signal. Any means by which the original signal might be transmitted to the measurement location would be likely to be subject to similar distortions as the link under test.
The present Applicant's co-pending International Patent applications WO96/06495 and WO96/06496 (both published on 29th February 1996, and U.S. Pat. No. 5,940,792) propose two possible solutions to this problem. WO96/06495 (now also U.S. application Ser. No. 08/765,697) describes the analysis of certain characteristics of speech which are talker-independent in order to determine how the signal has been modified by the telecommunications link. It also describes the analysis of certain characteristics of speech which vary in relation to other characteristics, not themselves directly measurable, in a way which is consistent between individual talkers, and which may therefore be used to derive information abou

REFERENCES:
patent: 5715372 (1998-02-01), Meyers et al.
patent: 5799133 (1998-08-01), Hollier et al.
patent: 5867813 (1999-02-01), Pietro et al.
IEEE International Conference on Communications 87 -Session 33.3, vol. 2, Jun. 7-10, 1987, Seattle (US), pp. 1164-1171, Edmund A. Quincy, "Prolog Based Expert Pattern Recognition System Shell for Technology . . . ".
Patent Abstracts of Japan vol. 17, No. 202 (E-2353), Apr. 20, 1993 & JP,A, 04 345327 (Nippon Telegr & Teleph Corp), Dec. 1, 1992.
IEEE Pacific RIM Conference on Communications, Computers and Signal Proocessing, Jun. 1-2, 1989, Victoria (CA), pp. 208-211, R.F. Kubichek et al, "Speech Quality Assessment Using Expert Pattern Recognition Techniques".
K. K. Paliwal and B. S. Atal, "Efficient Vector Quantization of LPC Parameters at 24 Bits/Frame," Proc. IEEE ICASSP, pp. 661-664, Apr. 1991.

Affiliated with

Also associated with

Rating

Trained artificial neural networks using an imperfect vocal trac does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Trained artificial neural networks using an imperfect vocal trac, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Trained artificial neural networks using an imperfect vocal trac will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-372210