Data processing: speech signal processing – linguistics – language – Speech signal processing – Psychoacoustic
Reexamination Certificate
1999-01-04
2001-06-05
Korzuch, William R. (Department: 2641)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Psychoacoustic
C704S232000, C381S094300
Reexamination Certificate
active
06243671
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to a device for analysis and filtration of sound. The invention also relates to a method for analysing and filtering sound.
BACKGROUND OF THE INVENTION
In many fields, it is important to be able to analyse sound in order to classify and identify different types of sound. For example, in the car and aviation industry it is often necessary to classify sounds from machines and engines in order to diagnose possible problems, such as cracks or worn bearings in the machine or engine. Speech recognition is another field where reliable analysis of sound is required. Today, a few different methods are employed for speech identification. A device for speech recognition is described in U.S. Pat. No. 5,285,522. Here, neural nets are used to recognise different voice patterns. The input signals to the neural nets come from a set of bandpass filters which separate the input acoustic patterns into frequency ranges. Another system for speech recognition is disclosed in U.S. Pat. No. 5,377,302, In this case, too, a neural net is employed, whose input signal consists of a signal filtered in multiple stages. The object of that system is to improve recognition of phonemes from speech. Both systems have difficulty analysing sound signals since they are not capable of distinguishing the magnitudes of the signals which are relevant to sound analysis. The reason for this is mainly that the processing of the signal prior to the neural filter cuts out important information and that the degrees of freedom are too limited. In addition, there are systems which work with neural nets only. However, in such systems, the neural nets cannot perform their task because the degrees of freedom are too many. It has become apparent that a certain amount of processing of the signal is required prior to it being fed to a neural net. Consequently, the neural nets do not have the intended effect in the known systems since they receive either too much or too little information about the signals. The same drawbacks as the ones associated with speech recognition would arise if one were to attempt to use the systems mentioned above for classification of machine or engine sounds.
The method most commonly employed today for classifying such sounds consists of simply listening to the machines with a stethoscope. Often, an experienced and trained ear has proven to be the most reliable tool for classifying and identifying various malfunction sounds and dissonances in, for example, an engine. Obviously, this method of classification has a number of drawbacks. Relying on a person's hearing is, of course, risky in itself since, for one thing, our hearing changes with time. In addition, it takes a long time for a person to bring her ability to recognise sound up to a level where misinterpretations are avoided to the greatest extent possible. Consequently, it is obvious that there is a need for a device which obviates the above-mentioned drawbacks of the prior art and which can analyse and filter sound in order to identify and classify accurately and reliably different sound types and sound patterns.
OBJECTS OF THE INVENTION
An object of the present invention is to provide a device for analysis and filtration of sound which unlike prior art devices permits reliable classification and recognition of different sound types and sound patterns.
Another object of the invention is to provide a reliable method for analysing and filtering sound.
SUMMARY OF THE INVENTION
According to the invention, a device and a method according to claims
1
and
7
have been provided in order to achieve the above-mentioned objects. The advantage of the invention is that the two different types of filtration, prior to the non-linear weighting, extract the information, such as magnitudes or eigenvalues, contained in an input signal which is relevant to the analysis and identification of sound and which provides the best conditions for efficient non-linear weighting.
One factor which contributed to the solution according to the invention is the insight that a system for the analysis of sound should emulate the manner in which our hearing analyses sound, something which will be set forth below.
According to a preferred embodiment of the device, the weighting means consists of a neural net. The filter combination is particularly suited for use with a neural net, which is thereby provided with a signal with a suitable number of degrees of freedom.
As mentioned above, to date there is no prior art which even remotely approaches the ability of the human ear when it comes to analysing the sound from, for example, an engine. Moreover, there is no known system which can compete with human hearing when it comes to identifying a particular voice. However, it is a requirement of sound identification that the person listening has “learned” the specific sound, i.e. the neural system and the brain must be able to relate the sound to previous experience in order recognise it. Consequently, the invention is based on the insight that in order to best analyse sound one must emulate an optimal, fully-trained human ear with its associated neural system. In order to analyse sound in a reliable manner, one must above all come to the realisation that the neural system which is responsible for evaluating the signals which are fed to the ear cannot be considered linear. On the contrary, our hearing reacts in a non-linear manner to changes of frequency, for example. Consequently, the present invention is also based on the insight that human hearing should be considered a non-linear system.
REFERENCES:
patent: 4959865 (1990-09-01), Stettiner et al.
patent: 5285522 (1994-02-01), Mueller
patent: 5377302 (1994-12-01), Tsiang
patent: 5450522 (1995-09-01), Hermansky et al.
patent: 5732388 (1998-03-01), Hoege et al.
patent: 5842162 (1998-11-01), Fineberg
patent: 4207728A1 (1993-09-01), None
Zhang et al., “Resonator based logarithmic filters,” IEE Proceedings I, Communications, Speech and Vision, vol. 139, Issue 1, Feb. 1992, pp. 36 to 44.*
Waters et al., “Least squares IIR filter design on a logarithmic frequency scale,” IEEE International Symposium on Circuits and Systems, vol. 1 May 1993, pp. 635 to 638.*
Strope et al., “A model of dynamic auditory perception and its application to robust speech recognition,” Conference Proceedings, 1996 IEEE International Conference on Acoustics, Speech, and Signal Processing, vol. 1, May 1996, pp. 37 to 40.*
Strope et al., “A model of dynamic auditory perception and its application to robust word recognition,” IEEE Transactions on Speech and Audio Processing, Vol. 5, Sept. 1997, pp. 451 to 464.
Lagö Thomas
Olsson Sven
Korzuch William R.
Lerner Martin
Smith , Gambrell & Russell, LLP
LandOfFree
Device and method for analysis and filtration of sound does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Device and method for analysis and filtration of sound, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Device and method for analysis and filtration of sound will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2469362