Data processing: speech signal processing – linguistics – language – Speech signal processing – Application
Reexamination Certificate
1998-08-20
2001-06-12
{haeck over (S)}mits, T{overscore (a)}livaldis I. (Department: 2741)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Application
C704S256000, C379S093110, C455S039000, C370S351000
Reexamination Certificate
active
06246985
ABSTRACT:
DESCRIPTION
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention generally relates to receiving data of different origins (voice, text, images, etc.) via communication channels. More specifically, the invention relates to using Hidden Markov Model (HMM) based methods to segregate data of different origins that were sent over a communication channel after reaching a user computer at a receiving end.
2. Background Description
In modern communication it is necessary to segregate voice input from fax, TTY and other non-voice data in order to either activate relevant devices (ASR, fax machine, TTY, phone mail etc.) or to activate relevant application programs in a user computer (e.g. display texts, pictures, etc.). For some kinds of data (e.g. fax and telephone voice data) segregation is done in a specially designed modem that uses a special protocol to establish where to transmit data (e.g. to fax or to a telephone).
When data reaches a destination end of a communication channel and enters a computer via a modem it passes through several Open Systems Interconnection (OSI) levels until it reaches an application level where it is used by an application program (e.g. data transfer, e-mail, user authentication etc.) (See reference: Andrew S. Tanenbaum “Computer networks”, 1988, Prentice Hall. N.J.). This data can be processed by an application program if the data is compatible with the application program's format.
An important example of processing multiple kinds of data are car embedded systems that may provide the following data trafficking: process data from satellites about a car location, process cellular telephone calls, accept credit cards by using cellular radio-based communication to verify credit transactions (in taxi cabs), send data to a car radio, process multimedia web data from Internet providers (to a portable car computer display), etc.
PROBLEMS WITH THE PRIOR ART
There are a number of problems with delivering data to a destination device. Some data may have protocols that are not compatible with protocols that are used by some OSI layer (e.g. by a modem at a physical layer or by an application level protocol) at a receiving end and therefore it cannot be delivered to a specified device automatically. For example, old fashioned TDD devices rely upon a person answering a phone to manually place a telephone pad on a TDD device when high pitched tones are heard (for a definition of TDD see L. Bahl, S. V. Degennaro, P. V. de Souza, D. Kanevsky and D. Nahamoo, “Communication aid for the hearing impaired based on automatic speech recognition”, IBM Technical Disclosure Bulletin, Y0888-0827, Vol. 32, No, 7 December 1989, ASR).
Currently, some organizations are attempting to develop standard protocols for data trafficking (for example, EIA TR30.1 and CCITT SGXVII WP1 are attempting to develop a standard for the modulation means of the asymmetrical modem). But individual data communication equipment (e.g. modem) manufacturers want to create a de facto standard for their own products and this prevents introduction of a single standard. Another reason that prevents introduction of a single standard is the fast growing number of different kinds of transmitted data and devices that process this data. For example, some biometrics data, that is extracted and transmitted to biometrics servers for user authentication, does not fall easily into earlier classification categories of textual or voice data. (See reference on standard issues in John A. C. Bingham,
The Theory and Practice of Modem Design,
John Wiley & Sons, New York, 1988).
Opportunities for automation in selection of applications to activate (e.g. display a picture, play an audio card, or to display a text) are similarly lost if an application layer is not familiar with data formats (or data delivering protocols), or a user wishes to send non-voice data without changing the type of communication, as is now the practice in a typical communication setting. Another example is mobile IP systems such as a networked car. The server must be redialed every time a different type of data must be transmitted. Alternatively more than one phone communication must be active in order to simultaneously have a user able to speak with a car assistance operator, enable the operator to read about the car, enable the user as well as the operator to know the exact position of the car, etc.
It would be convenient if there were only one phone communication and the different functions could be activated sequentially.
SUMMARY OF THE INVENTION
It is therefore an object of the present invention to provide an improved system and method for segregation of different type of data at a receiving end in a communication channel.
A further object of the invention is a system and method that provides hierarchical segregation of different types of data along a communication path at a receiving end of a communication channel, in order to send different kinds of data to different destination devices.
Another object of the invention is an improved system and method that uses hidden Markov models that are trained on different kinds of input data to provide segregation of this data at a receiving end in a communication channel.
It is also an object of invention to have a system that detects changes in a type of input data and matches segments of data of different types with stored data prototypes to define the nature of data.
Still another object of the invention is a system that detects changes in type of input data using likelihood ratio and matching segments of data of different types with HMM based stored prototypes, in order to define the nature of data.
A further object of this invention is an improved system and method that uses linear hidden Markov models that are trained on different kinds of input data (voice, textual, pictorial, animated data) to provide segregation of this data at a receiving end (computer, CPU, embedded system in a car), in a communication channel (over a telephone, over a network, ISDN, dedicated lines, via radio, a satellite communication system).
In accordance with the invention, a Hidden Markov Model is used to train on different kinds of input. For example, a linear HMM consists of states that are connected with arcs and loops (arcs that connect states with themselves). Each arc and loop give rise to probability distributions that are modeled by a mixture of Gaussian distributions. This HMM can then train on the following data that is sent via different types of communication channels (telephone, digital, via radio, delicate lines etc.): 1) a wide variety of telephone speech, 2) TTY signals, 3) Baudot signals, 4) fax signals, 5) pictorial data, and 6) animated data.
The length of the sequence of HMM imposes a minimum duration. The invention begins by finding the best path alignment at the beginning of a signal. Acoustic data can be represented as a string of frames, each frame corresponding to a vector that stores some acoustic information for some moment of time. This acoustic data segment (a string of frames) approximately corresponds (in duration) to HMM set of states. There is a standard procedure that permits alignment of the HMM set to a string of frames. Different HMMs would produce different alignments to this acoustic string. The HMM best matched to the acoustic string would give the highest alignment score. Therefore, defining the best aligned HMM allows definition of the most probable acoustic source. This will automatically determine the nature of the signal. Then a module that recognizes data by type is located before a block that sends data to different devices that are connected to a receiving communication end or to different application programs inside a particular device.
After the data type is defined it can be either sent to a designated device (e.g. fax machine, telephone, TDD, PC, prints etc.) or to a relevant application program (e.g. editor for textual data, Automatic Speech Recognition module for speech, an image editor for images, music editor for music, etc.).
Data is segreg
Kanevsky Dimitri
Maes Stephane H.
Zadrozny Wlodek Wlodzimierz
Zlatsin Alexander
International Business Machines - Corporation
McGuireWoods LLP
Percello, Esq. Louis J.
{haeck over (S)}mits T{overscore (a)}livaldis I.
LandOfFree
Method and apparatus for automatic segregation and routing... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for automatic segregation and routing..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for automatic segregation and routing... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2499000