Call routing based on caller's mood

The present invention relates to the telephone field and especially intelligent call processing devices and a method for routing calls.

In the business world it is becoming increasingly common, so-called Call center set up for communication with customers to orders to accept and handle customer complaints. Frequently these automated private branch exchanges (PBX; Private Branch Exchange) is used, they should submit the customer choices and the call to the customer's reactions to an appropriate lead operator.

From the <patcit><text>US 5,553,119</text></patcit> For example, a switching system to connect a call between a calling station and a called station known to the one to a system and a method having voice recognition, which of a plurality of in a demographic Database stored preset responses and reactions are the most appropriate response to the caller's response.

In the US patent 5,493,608 is proposed, the playback speed of a response system adaptively adjust to the reaction rate of the caller, to the processes for the user to optimize when this is already an experienced user.

The aforementioned systems from the prior Technology all point to a problem that no statement and line the caller according to his attitude or mood possible is and the care in this respect has characterized significant weaknesses.

The problem with the existing systems lies in that, while many calls from everyday Kind are such as simple technical objections, and thus can be handled by trained editors, some calls, eg from angry Customer complaints, a treatment of especially trained require processors. It is important that such calls usefully be treated to prevent a loss of fiscal order. When present system, an operator of the caller Tell hand at a particularly trained arranger. This is not only inadequate, the delay, resulting from the handing over gives to another operator, can a sensitive angering customers often, the already not anyway in high spirits. It is generally for the Caller or the caller desirable to know that he or it receives special treatment.

The object of the invention is to provide this mitigate problem.

o According to the present invention a Rufleitverarbeitung is provided, the answering device a, an interactive voice unit for eliciting information about the Call from the customer, a behavioral response means for analyzing the reaction of the caller according to previously established criteria to identify callers certain have typical behavioral characteristics, and a guide for Directing callers who are identified, these typical behavioral characteristics to have, contains at predetermined operator stations.

A typical behavior characteristic can be the mood of the caller. For example, angry callers to particularly are routed trained agents.

The analyzing means can for example a speech recognizer, the high tension or anger detects a touch-tone timing device, which the impatience or annoyance the caller detects when entering this Tastwahlimpulse, a Speech recognition means that the sex and approximate age the caller (male, female, child detected), and a speech recognizer be, the typical language features such example accents or stuttering detected.

With the device according to the invention may call center much more powerful be. By automatically detecting the mood and switching the call to agents that this in the treatment of calls Art is good, the processing smoother, faster and with higher customer satisfaction handled. The call can be treated by that operator be, that is best for the treatment of this type of calls, or the operator could adapt their style to meet the needs the caller comply.

The invention provides for conducting Calls in a call center also provides a method, the following comprising steps of: Receiving an incoming call in an interactive voice unit, Analyzing the response of the caller in accordance with predetermined criteria, to identify typical behavioral characteristics, and directing the to an agent that is identified according to the call selected phenotypic characteristics is.

The invention will now be detailed going by way of examples only with reference to the accompanying drawings, in which:

<figref idrefs="S13">1</figref> on Block diagram of a system ge&lt;?page 3?&gt;according to the invention is

<figref idrefs="S13">2</figref> on Block diagram of a touch-tone analyzer is,

<figref idrefs="S14">3</figref> on Block diagram of a mood detector,

<figref idrefs="S14">4</figref> on Block diagram of a word detector, and <figref idrefs="S15">5</figref> a block diagram of a detector is typical of speaker characteristics.

A switching device <figref>1</figref>. eg an automated branch exchange (PBX; Private Branch Exchange) of Mitel Corporation is on incoming leads <figref>2</figref> connected. contains the PBX an answering machine unit <figref>13</figref> and an interactive voice response unit (IVR; Interactive Voice Response) <figref>3</figref>, The interactive voice response unit<figref>3</figref> allowed there a caller options in response to calls by using a touch-tone keypad to choose.

The one lead unit <figref>14</figref> containing switch <figref>1</figref> forwards the calls on lines <figref>10</figref> to Call agents. Each call agent station is connected to a computer<figref>16</figref> and a telephone <figref>15</figref> provided.

The switch <figref>1</figref> is via an output line <figref>9</figref> at connected a Stimmungsanalysierer. The Stimmungsanalysierer consists of a Tastwahlcharakteristik unit <figref>5</figref>, A mood detector <figref>6</figref>. a word detector <figref>7</figref> and a speaker characteristic unit <figref>8th</figref>,

The Tastwahlcharakteristik unit <figref>5</figref> revising the time by touch-tone shades, the reaction rate and the frequency at which the selection takes place. For example, a very generous selection indicative for one his customer who upset or is in a hurry. Repeatedly pressing the same button is also a sign of anger or impatience. an extended To press a Tastwahlknopfes may be a sign of an angry his customers. The Tastwahlcharakteristik unit<figref>5</figref> analyzes the Tastwahlsignale a neural network device and provides an output signal ago that the level of potential impatience of the caller or the anger of the caller on the basis of the aforementioned measurements indicative. This output signal becomes a Stim-drying logic unit <figref>4</figref> sent.

Suitable are, for example neural networks in the article "An Introduction to Computing with Neural Nets ", IEEE ASSP Magazine, pages 4 to 22 April 1987.. Neural Networks have the ability to on the basis of past experience to "learn".

The mood detector <figref>6</figref> monitors the caller's voice and analyzes sounds or other typical Voices features, such as speech rate, volume, or Response speed with a neural network device. On the Based on these characteristics determines the mood detector unit <figref>6</figref> the Level of potential irritation or the impatience of the caller. This parameter is also the mood logic unit<figref>4</figref> supplied.

The word detector <figref>7</figref> listens to the Voice of the caller and performs keyword recognition specific words from that annoyance an or impatience would show. These are, for example, swear, curse words or Additional rates Show the impatience (eg 'now join already "). A speaker independent Speech recognizer taps from these features and feeds it to a neural network device, the potential annoyance or impatience of the caller calculated. A suitable device for this is as BI and PD Pawate Robinsons in "Implementation of an HMM-Based Speaker-Independent Speech Recognition System on the TMS32OC2x and TMS32OC5x ", Texas Instruments Incorporated, described 1996th This parameter we Mood logic unit<figref>4</figref> supplied.

The speaker characteristic unit <figref>8th</figref> monitors the caller's voice and analyzes the speech sounds and other typical specifications, such as the frequency spectrum, for determining the approximate Age and sex of the caller. The pitch is with a pitch detector recovered. The logic in this block determines whether the speaker is a male Adult, female adult or a child. They also checked Sounds to determine whether the caller has a strong accent which a potential language preference displays. This information is the mood logic unit<figref>4</figref> supplied.

The Mood logic unit <figref>4</figref> includes logic such as a neural network means that all available information combined to the potential degree of irritation of the caller, the degree of impatience, if the caller male or female or a child, and a potential language preference to determine. This information is the PBX or a Analog / digital converter fed, before the call to an agent by the guide <figref>14</figref> in the switching device <figref>1</figref> is switched.

The information is in two Way used: If the annoyance level exceeds a preset threshold, by the mood logic unit <figref>4</figref> is determined is they used by the switch to transfer the call to Group to conduct of agents, the fretting in dealing with Customers are good. The sex of the caller can be used, the to conduct calls to agents of the appropriate gender to the treatment improve the caller. The level of potential or impatience annoyance, as the mood logic unit <figref>4</figref> was determined, is then displayed on the PC screen <figref>16</figref> represented the agent, to the potential level of the mood of the caller display. The agent This then can be used for use the Opening Welcome and confirmation this mood from dealing with the caller expect listening. &lt;?page 4?&gt;If sentiment indicator was inaccurate, the agent can rapidly to a recourse normal treatment of the call.

The degree of accuracy, in the System by the agent along the line <figref>11</figref> fed back be, the weights in the neural network means be brought up to date, which improves the accuracy for future calls increases.

The Tastwahlcharakteristik analyzer is detailed in <figref idrefs="S13">2</figref> shown. The signal from the line <figref>9</figref> is a period of time analyzer <figref>20</figref>. determines the Tastwahldauer, and a DTMF analyzer <figref>22</figref> supplied to the the down depressed Key identified. A decision unit<figref>24</figref> determines whether it is the same button as the previous button. The outputs the decision unit <figref>24</figref> and the length of time-analyzer <figref>20</figref> will a touch-tone analyzer <figref>26</figref> supplied, the annoyance level to of these parameters on the basis of previous experience may determine.

The mood detector <figref>6</figref>. the Indian <figref idrefs="S14">3</figref> detailed As shown, leads the signal from the line <figref>9</figref> a sound analyzer <figref>30</figref>. a speech analyzer <figref>32</figref>, The rate of speech identified the caller, a volume analyzer <figref>34</figref> and a Answer Time Analyzer <figref>36</figref> to. These four individual analyzers is in a mood detector analyzer <figref>38</figref> on Output created.

A word detector <figref>7</figref>, of the in detail in the <figref idrefs="S14">4</figref> shown is, contains a keyword database <figref>40</figref>. a speech recognizer <figref>42</figref> and a word parser <figref>44</figref>, When a forward speech <figref>42</figref> unidentified word in the Database <figref>40</figref> is found, the word parser <figref>44</figref> on Output to logic mood <figref>4</figref> from that of the nature of the word dependent is. Some words clearly show a greater degree of annoyance than others.

The speaker characteristic unit <figref>8th</figref>. in more detail under the <figref idrefs="S15">5</figref> shown is, contains a sound analyzer <figref>50</figref>, A frequency spectrum analyzer <figref>52</figref>. a pitch analyzer <figref>54</figref> and a Sprecheranalysierer <figref>56</figref>, The Sprecheranalysierer<figref>56</figref> provides an output signal, an indication of the age, sex the accent, the language and so the caller is.

When a call from the call center switch <figref>1</figref> receive is, he is first with the interactive voice response unit <figref>3</figref> and the Stimmungsanalysierer connected. The interactive voice response unit<figref>3</figref> asks the caller to provide information such as the account number of the caller or to the type of call s (eg reservation, flight arrival, invoice issue) usw.Während this information in a conventional be collected and processed way, the same speech samples and Tastwahleingaben in terms of the mood of the caller by the units <figref>5</figref> to <figref>8th</figref> analyzed. The call is then the next available directed agent who for the treatment of this type of call is good. For example, a to fretting male / female person to a patient female / male agent are passed, the been trained in dealing with angry customers is. A caller who is in a hurry, is a trained agent passed, and a message indicating this is on the screen of Agents displayed instantaneously. The call is then to normal treated manner.

Error in the mood detection be quickly corrected when the agent perceives that the automated Display is incorrect, and he then accesses a normal settlement back. The agent can the degree of inaccuracy into the system via conduit <figref>11</figref> rückkoppeln, whereby the accuracy of future calls is increased. Database<figref>12</figref> prestores collected Data that can be used to the accuracy of the system to increase. The mood detection devices can many Parameters measured such as: signs of tension in the language, very spacious Answering to questions added words in the language, the moods View (eg swear words), only occurring in other languages ​​sounds that a sign of a Accent are poor automatic speech recognition, either an accent or a Solecism indicating rapid pressing touch-tone buttons or very long press by touch-tone buttons, what anger indicating multiple presses a touch-tone button, with only one press is required.

The results of all these measurements are combined to the seemingly appropriate customer mood parameters to calculate. This parameter is then used by the switching device, the call to the most appropriate available agent to conduct, and the parameter is also quickly on the screen the agent indicated to treat for to provide an indication of the likely condition of the caller.

Earlier customer data to Mood / voice parameters are added. For example, the Database <figref>12</figref>After a customer or a customer's or their account number entered, contain information to the effect for example, that this Customer and this customer frequently is in a particular mood or recently had a problem or a special typical speech feature has. If the tuning screen was correct, could this data is added to the customer data, which appears is that the Mood display mood of customers has measured appropriately, and it could a past record of customer transactions are conducted, what comes in good stead prediction on subsequent calls.

Mood displays can be at use many applications that include processors of mechanisms, the inclination of the editor &lt;?page 5?&gt;to detect errors.

Mood ads can use up to a frustrated caller who is incapable of proper selection in a Sprachpost- (voice mail) system to meet or interactive voice response system, disengage. This party could For example wiedefiolt press "0", in hoping to reach a human being. This process could be detected and suppress the normal Leitvorgang to to convey the caller to an agent.

The invention thus provides against the known systems for a considerable improvement, which consists in the fact that they Call possible to effectively to be directed to a staff that qualifies fittest is to treat an incoming call.