WO2007073798A1

WO2007073798A1 - Call routing based on context sensitive speech recognition

Info

Publication number: WO2007073798A1
Application number: PCT/EP2006/010595
Authority: WO
Inventors: Wouter Johan Peter Van Toledo
Original assignee: Koninklijke Kpn N.V.
Priority date: 2005-12-27
Filing date: 2006-11-06
Publication date: 2007-07-05

Abstract

The application relates to call routing in a call center context. Known call routing solutions comprise the detection of DTMF signals, as issued by the user or the recognition of one or more keywords as spoken by the user. Currently available routing solutions for service calls based on the detection of DTMF signals or keyword recognition offer either a limited set of options to the customer from which the customer can choose or a limited set of keywords that can be recognized. The invention is based on the insight that recognition of questions or problem descriptions from users seeking service show resemblance at a conceptual level with text searching in unstructured data, more recently, a new system for text searching in unstructured data has been developed and the invention is to be seen in the application of this new system to the call center context as mentioned above .

Description

Title

CALL ROUTING BASED OW CONTEXT SENSITIVE SPEECH RECOGNITION

Field of the invention

The invention relates to call destination generation and call handling. More specifically the invention relates to the generation of a call destination based on voice recognition and the subsequent call handling.

Background of the invention

Known call routing solutions in a call center context comprise the detection of DTMF signals, as issued by the user or the recognition of one or more keywords as spoken by the user. Based on these DTMF signals or one or more keywords, known routing solutions decide to which destination a service call should be routed, i.e. which type of call center agent should be able to service the user . More recently, a context sensitive voice recognition system has been developed, as described in http : //www. autonomy. com/downloads/White%20Papers/Autonomy%2 0White%20Papers/Autonomy%20Technology%20WP%2020040105.pdf. This system enables the recognition of naturally spoken sentences. This allows the matching of contexts or concepts, rather than only the recognition of keywords.

Problem definition

Currently available routing solutions for service calls based on the detection of DTMF signals or keyword recognition offer either a limited set of options to the customer from which the customer can choose or a limited set of keywords that can be recognized. Both methods result in an error margin, i.e. a number of faultily routed calls. These faultily routed calls have to be re-routed to the appropriate call center agent by the call center agent who initially receives the call. The faultily routed calls result in additional handling time and routing time, which lead to extra costs for the provider of the call center service and extra delay in the service for the user.

Aim of the invention

It is an object of the invention to reduce the problems identified in the prior art. It is a further object of the invention to further improve the experience for the user of a call center service.

Summary of the invention

The current invention provides a method and system for generation of a call destination as a function of voice recognition and the subsequent call handling.

According to an aspect of the invention, a call destinations is generated as a function of voice recognition, whereby the voice recognition comprises pattern matching using predictable statistical word patterns representing concepts. According to this aspect of the invention as input a spoken sentence comprising a problem description or question is recognized, this recognizing resulting in a single one concept or more than one concepts, followed by the selecting of the single one concept or the concept with the highest contextual relevance when the recognition resulted in more than one concepts, followed by the relating of this selected concept to a call destination and the subsequent generating of the related call destination as output.

According to a further aspect of the invention, the generation of a call destinations as a function of voice recognition comprises the receiving of feedback from a call center agent, this feedback comprising the sentence as recognized in text format and the correct call destination, followed by updating the database or model by relating the concept corresponding with the sentence as recognized to this correct call destination.

According to another aspect of the invention, the system disclosed by the invention generates call destinations as a function of voice recognition comprising pattern matching using predictable statistical word patterns representing concepts and this function comprises the relating of the concepts to call destinations.

According to a further aspect of the invention the system stores concepts and call destinations in a database or a model.

According to a further aspect of the invention the system forwards the sentence as recognized to a call center agent in text format.

Brief description of the drawings

The invention will be explained in greater detail by reference to the drawings, in which:

Fig. 1 shows a schematic representation of an embodiment according to the invention. Fig. 2 shows a schematic representation of relations between concepts and call destinations.

Detailed description of the invention

For the purpose of teaching of the invention, exemplary embodiments of the method and system of the invention are described in the sequel. It will be apparent to the person skilled in the art that other alternative and equivalent embodiments of the invention can be conceived and reduced to practice without departing from the true spirit of the invention, the scope of the invention being only limited by the claims as finally granted.

According to the current invention, the more recently developed system for context sensitive voice recognition is used for generating a call destination as a basis for the routing of service calls to the appropriate call center agent. The context sensitive voice recognition allows the user seeking service to state the service that he seeks, or the problem he is experiencing, in a general sense. For example, the user can express a question like "I want to increase the bandwidth of my ADSL line". Because context sensitive voice recognition recognizes contexts or concepts rather than only keywords, the percentage of faultily routed calls will be much lower than in the case of keyword recognition .

The invention is based on the insight that recognition of questions or problem descriptions from users seeking service show resemblance at a conceptual level with text searching in unstructured data. The system according to the current invention generates call destinations as a function of context sensitive voice recognition. This is illustrated in figure 1, in which the input to the system (2) is formed by a naturally spoken sentence by a user seeking service. The naturally spoken sentence contains a question or problem description, such as "The quality of the channels on my TV is very bad" or "How can I increase my ADSL bandwidth?". The naturally spoken sentence is modulated or digitized according to common telephony techniques, such as with POTS or ISDN. The voice recognition unit (2' ) receives the signal and processes the naturally spoken sentence from the user. The processing by the voice recognition unit involves interaction with a model or database (4) which contains concepts, with the aid of a central unit (6), which is executing a program (7). The processing results in one or more recognized concepts, managed by the program (7). In case more than one concept is recognized, these concepts are ranked according to their contextual relevance. The single concept that was recognized or the concept with the highest contextual relevance is selected by the program and the related call destination is selected by the program.

Figure 2 illustrates the relation between concepts and call destinations as maintained in the database or model. As illustrated, concept A, C and F are related to call destination 1, which means that the selection of either concept A, C or F will lead to relating the selected concept to call destination 1. The selection of concept B or H will lead to relating the selected concept to call destination 2. The program (7) instructs the generator (3') to generate the call destination related to the selected concept as output signal (3) . In practice, the output of an embodiment of the invention is connected to a system for call termination in an Intelligent Network platform. This allows the service call to be forwarded to the appropriate call center agent, corresponding with the generated call destination. In addition to the generation of a call destination, the system according to the invention is able to forward the sentence as recognized in text format to the call center agent operating the call destination by means of a Call Telephony Integration interface. The sentence as recognized in text format can be displayed on the call center agent's PC screen, which enables the call center agent to better service the user seeking service.

The embodiment according to the current invention supports a "self training mode", in which a feedback (8) is connected from the call center agents to the system (1) in figure 1. In the "self training mode" the system can receive input from call center agents via the feedback input (8), the information in the feedback indicating that a call destination was incorrect and which call destination should have been generated by the system. To this end the call center agent sends both the sentence as recognized in text format, the wrongly generated call destination and the correct call destination to the system input (8). The feedback input (8) operates as a common CTI interface, allowing text messages to be sent to the system. In the

"self training mode", the system will update the database or model (4) according to the feedback received from the call center agent. After receiving the feedback, the system updates the database or model (4) such that the concept corresponding to the sentence as recognized in text format (in the feedback) is related to the correct call destination, as indicated by the call center agent.

Claims

1. Method for generating call destinations as a function of voice recognition characterized in that said voice recognition comprises pattern matching using predictable statistical word patterns representing concepts and the method comprises the steps of

- recognizing as input a spoken sentence comprising a problem description or question and said recognizing resulting in a single one concept or more than one of concepts;

- selecting the single one concept or selecting the concept with the highest contextual relevance when the preceding step resulted in more than one concepts; - relating said selected concept to a call destination;

- generating said call destination as output.

2. Method according to claim 1 comprising the steps of:

- receiving feedback from a call center agent, the feedback comprising the sentence as recognized in text format and the correct call destination;

- updating the database or model by relating the concept corresponding with the sentence as recognized to the said correct call destination.

3. System generating call destinations as a function of voice recognition characterized in that said voice recognition comprises pattern matching using predictable statistical word patterns representing concepts and said function comprises relating concepts to call destinations .

4. System according to claim 3, wherein concepts and call destinations are stored in a database or a model.

5. System according to claim 3 or 4, wherein the sentence as recognized is forwarded to a call center agent in text format.