The invention relates to a method of making a connection for voice communication from a terminal to a destination subscriber device within a communications network, in which a first connection for voice communication is first made to a voice dialoging system that communicates with a user of the terminal by using an automatic speech recognition device and an information-emitting device, a subscriber identifier for the destination subscriber being obtained by means of a dialog between the user and the voice dialoging system, and a second connection for voice communication is then made between the terminal and the destination subscriber device by using the subscriber identifier that has been obtained. The invention also relates to an automatic voice dialoging system, to a terminal and to a system, comprising the voice dialoging system and the terminal, for making a connection for voice communication by the said method.

The classic manner in which a connection for voice communication is made between a terminal, such as a hard-wired telephone, a mobile telephone or some other device that performs telephonic functions, such as a suitably equipped PC, laptop or PDA (personal digital assistant), and a destination subscriber device is that the user enters manually a subscriber identifier, e. g. a telephone number, for the desired destination subscriber device.

For some time now, there have also been terminals on which it is possible, from a telephone directory stored on the terminal, to obtain the desired subscriber identifier with the help of manual inputs, e. g. by inputting or selecting a name. The terminal then dials the destination subscriber device, using the subscriber identifier that has been entered or obtained, either automatically or after receiving additional confirmation.

A relatively novel alternative to this is so-called voice dialing in which the subscriber identifier is obtained from spoken commands made by the user and the right connection is made in this way. The way in which the subscriber identifier may be obtained in this case is that the user speaks the individual characters or words of the subscriber identifier and a speech recognition device recognizes the characters or words and puts them together to form the desired subscriber identifier. What happens in many cases however, is that here too there is a memory or a database available that contains a"telephone directory", and the user simply gives the name of the desired destination subscriber, and if necessary

other information such as parts of the latter's address, and the correct subscriber identifier is found from these details. There are many advantages for the user in voice dialing of this kind, particularly in cases where the user is doing other things at the same time and needs both hands for them, such as driving a motor vehicle for example. What is more, voice dialing enables even disabled persons, who are for example blind or whose manual dexterity is severely restricted, to use a communications terminal of the appropriate kind and to call any desired other destination subscribers, without outside assistance. In addition to this, voice dialing is also unusually convenient for the user.

At the present time, there are two different ways in which voice dialing of this kind is implemented in practice.

In the first method, the voice dialing takes place in the communications terminal itself. In this case, a voice dialing function in the communications terminal is activated by pressing a button or by means of an activating word. With the voice dialing function activated, the user can say a name or other information. A speech recognition device arranged in the terminal itself attempts to recognize the name and to correlate it with an entry in the telephone directory stored on the terminal. If it succeeds in recognizing the name and in correlating the name correctly, the appropriate telephone number is dialed and the desired connection thus made. With this form of voice dialing, the list of names, i. e. the telephone directory, is compiled and updated in the terminal by being spoken in and/or by being entered via the keyboard. A terminal-based method of this kind has the advantage that it is simple to organize because all the provisions required can be decided on and made by the manufacturer of the terminal. There is no need for any assisting work to be done by a network operator, and particular there is no need for him to provide a switching infrastructure. What is problematic about this method however is that, because of their size and the allowable cost of manufacture, the computing power and storage capacity that communications terminals have are generally very limited. Because of the limited resources, relatively simple speech recognizers therefore have to be used. A further disadvantage of a terminal-based voice dialing method is that the terminal has to be configured specifically for the language of the country concerned, which is a burden for manufacturers, retailers and users.

The disadvantages mentioned can be alleviated or in some cases even completely avoided by a second procedure that employs a method of the kind mentioned in the opening paragraph. This is a network-based method of voice dialing because use is made in this case of voice dialoging systems arranged within the communications network to obtain the subscriber identifier. Dialoging systems that communicate with a user by using

speech-recognition and information-emitting devices have been known for quite some time now. These are voice-controlled automatic systems that are often referred to as voice portals.

An example of these are automatic call-answering and information-supplying systems of the kind that are now being used by, for example, certain large firms and authorities to supply a caller with the desired information as quickly and conveniently as possible or to connect him to a department that is responsible for dealing with his specific requirements. Another example of them are automatic directory-enquiries systems of the kind that are already being used by certain telephone companies. The voice dialoging systems may be dialoging systems that communicate with the user by speech alone, i. e. where all the data or commands from the user are entered in spoken form and are recognized in a speech recognition device so that the system can react to them in the appropriate way. Similarly, all the input requests or data going from the system to the user, i. e. all the system prompts, can be emitted by means of a suitable facility for spoken outputs, a so-called prompt generator. TTS (text to speech) converters for example may be used as prompt generators, or else prepared, stored sound files are read out. Voice dialoging systems of this kind may also be mixed dialoging systems (multimode dialoging systems) that emit the information going from the system to the user by means of a display for example, which the user is once again able to read without using his hands, with prompts too being given in some cases as speech outputs and in some cases via the display. Similarly, it is possible for command inputs given by key presses, using for example so-called DTMF (dual-tone multi-frequency) signals in the case of conventional telephones, to be accepted by the system as well. To communicate with the dialoging system, the communications terminal must first make a first connection for voice communication with the voice dialoging system. Once this first connection for voice communication has been made, the subscriber can enter the name of a desired subscriber in the same way as with the terminal-based method. The dialoging system can then obtain the desired subscriber identifier from a list of names, which is also used for speech recognition and which contains the correlations of the names, their pronunciations and the associated subscriber identifiers. A list of names or telephone directory of this kind may be user-specific, i. e. it may be a personal address book that is assigned to a given user. It may however also be a general telephone directory corresponding to a public telephone directory or a company telephone directory. It is even possible for personal name entries to be interpolated into a general telephone directory. The personal address book or the personal part of the address book may be updated in this case either by a spoken dialog or, where the voice dialoging system itself provides this facility, in textual form, e. g. over the internet. Once the subscriber identifier has been found,

the second connection for voice communication between the terminal and the destination subscriber device is made automatically within the network, i. e. the user is connected automatically. A method of this kind is described in US 5,297, 183 in which a speech recognition system sends the subscriber number that has been recognized to a mobile telephone switching exchange and thus causes the subscriber to be connected to the destination subscriber. This network-based method has the advantage that it is more versatile than the terminal-based method. Hence, far larger telephone directories can be created and telephone directories can be used from a plurality of terminals. Also, extensions of and improvements to the technology can be made at any time in a way that is transparent to the user. Furthermore, it is possible for considerably more complicated speech recognizers to be used that are capable of accomplishing recognition safely even under extreme conditions.

The disadvantage this method has is that greater demands are made on the infrastructure of the communications network. In particular, the network nodes have to be capable of performing or initiating the switching. For this it is necessary for assisting work to be done by the network operator. The SS7 protocol and the complicated and expensive infrastructure it involves are often required. Added to this is the disadvantage that the billing of the user is considerably more complicated because, when a call is switched forward, the charges for it have to be booked to the user's account. A change of tariffs is generally required for this purpose and the technical complication this involves is relatively high and the change will not be very clear to the user on the bill that he later receives.

It is therefore an object of the invention to provide a simple and inexpensive alternative to the method described that on the one hand enables a dialog to be carried on with the user in as convenient a way as possible and in particular ensures good speech recognition, and in which on the other hand connections are made easily and in a way that the user can control and monitor satisfactorily without any complicated and expensive demands being made on the switching infrastructure on the network.

In a method of the kind described in the opening paragraph, this object is achieved by virtue of the fact that the subscriber identifier that has been obtained is first transmitted to the communications terminal and the communications terminal automatically makes the second connection for voice communication using the subscriber identifier.

The method according to the invention combines the advantages of the two known methods and at the same time overcomes their respective disadvantages. In particular,

the method according to the invention on the one hand allows very complicated speech recognition and the use of telephone directories of almost unlimited size. General telephone directories that are regularly updated without any action having to be taken by the user may be used in this case. On the other hand, no additional complicated and expensive switching infrastructure is required within the network. Because two separate connections are made, a first connection for voice communication to the voice dialoging system and a second connection for voice communication, separate from the first connection, between the user's communications terminal and a destination subscriber device, there is no change of tariff, in contrast to a connection on the network. Billing is therefore unusually simple, and transparent and comprehensible to the caller.

To enable the method to be executed, automatic voice dialoging systems must have at least the following components: connecting means for making a connection for voice communication to a terminal; an automatic speech recognition device and an information-emitting device, e. g. a pure speech-output facility and/or a device for emitting machine-readable data for, for example, a display, e. g. text or graphic data, or for controlling a local speech-emitting device belonging to the terminal to enable the dialog to be conducted with a user of the communications terminal; a dialog-controlling device to control the dialog between the user and the dialoging system to enable a subscriber identifier to be obtained for a destination subscriber; -means for transmitting the subscriber identifier to the terminal in machine- readable form.

What the phrase"in machine-readable form"should be understood to mean in this case is that the subscriber identifier is not transmitted as a speech output but in a form in which, without any speech recognition, it can be automatically further processed and used by the terminal to make the second connection for voice communication. Furthermore, the subscriber identifier may, in addition, be emitted as a speech output to enable the user to check that the correct subscriber identifier is being dialed.

A suitable terminal for executing a method of this kind requires a speech user- interface that usually has a speech input device and a speech output device and/or a display.

It must also have means for making a first connection for voice communication with a voice dialoging system to enable the desired dialog to be conducted to obtain the subscriber identifier. In accordance with the invention, it must also have means for receiving the

subscriber identifier that is obtained in machine-readable form and for automatically making a second connection for voice communication to the desired destination subscriber using the subscriber identifier. What this means is that the terminal must have a suitable device for automatically dialing the destination subscriber device using the subscriber identifier it has received.

A system for executing the method according to the invention must have at least a voice dialoging system of this kind and one of the above-mentioned terminals, which can be connected together via a communications network.

The communications network may be a communications network of any desired type. In particular, it may even be a network assembled from a wide variety of individual networks such as a GSM network, a hard-wired network, a UMTS network, the internet, etc. What is essential is simply that a connection for voice communication can be made between the devices and terminals concerned. As a function of the particular networks to which the devices and terminals are connected, the types of subscriber identifier involved may even differ, being for example a telephone number, a URL or some other address code.

Other advantageous embodiments are specified in particular in the dependent claims.

In an advantageous variant, on receipt of the subscriber identifier, the terminal first makes an enquiry for confirmation to the user. Only if a confirming command is received from the user is the second connection for voice communication then made. This has the advantage that the user has considerably better control over what connections are actually made, which is important in that every making of a connection involves a charge to the user.

It is possible in principle for the first connection for voice communication to the dialoging system to be maintained while the second connection for voice communication to the destination subscriber device is being made. For this to be the case, it is merely necessary for the terminal to be capable of maintaining two connections for voice communication in parallel with one another. Further recognition and identifying processes can then be carried out by the voice dialoging system in the same way, e. g. a further subscriber identifier can be obtained for a different destination subscriber and transmitted to the terminal. The communications terminal can then initiate further connections. This can be done simultaneously with the second communication connection (for three-person conferencing) or subsequently thereto.

In the majority of cases however, the terminal will, on receipt of the subscriber identifier, terminate the first connection for voice communication either automatically or on receipt of a command to disconnect from the user. It is also possible in this case for the communications terminal first to emit a"Disconnect"query to the user, i. e. a suitable prompt that queries whether the connection to the dialoging system is to be terminated. This may for example take place before the query as to whether the second connection for voice communication is to be made.

There are various possible ways in which the subscriber identifier can be transmitted to the communications terminal. On an analog network, on an ISDN network or on a GSM network, it is sensible for the subscriber identifier to be transmitted in acoustic form, encoded via the particular connection for voice communication. What this means is that a coding is applied with which the data can be transmitted over the speech channel. An example of such coding is DTMF coding.

On networks that allow a speech channel and a data channel to be connected simultaneously, such as a GPRS network for example, a second data channel may be connected as well and used to send the subscriber identifier directly in the form of suitable data. With certain networks, e. g. UMTS networks, or in the case of internet telephony, i. e. in the case of an IP connection, it is also possible for the subscriber identifier to be sent directly as data of the appropriate form, because data and speech packets can be transmitted mixed over a single channel on these networks However, even on networks on which it is possible for data to be transmitted in a usual format, it may under certain circumstances be better for the subscriber identifier first to be coded into a form that can be transmitted over the speech channel. This is true particularly where, as in the case of GPRS, additional costs are incurred as a result of the connecting up of a data channel.

A further possibility is for the subscriber identifier to be sent via a short message service, such as SMS. This, however, may cause delays in the dialing operation because messages are not always delivered immediately in short message services.

Alternatively or in addition to voice dialing by name with associated subscriber identifiers, a direct form of voice dialing as defined in claim 5 may be used. In this case, the user speaks the desired subscriber identifier, such as the telephone number or URL, directly as figures or natural numbers or as words or letters. This speech input is recognized by the dialoging system. The outcome of the recognition is then sent back to the user's communications terminal in the form of machine-readable data, i. e. in for example the form

of a sequence of numbers or letters, and the communications terminal in turn automatically makes the second connection for voice communication in the usual way using the outcome of the recognition, i. e. the subscriber identifier that has been obtained.

In a particularly convenient embodiment dealt with in claim 6, the dialoging system is assisted by one or more operators, at a call center for example. In this case, the voice dialoging system is so arranged that in certain cases, rather than being obtained by automatic speech recognition, the desired subscriber identifier is obtained by human operators and is sent back to the terminal in the usual way. It is sensible for this passing on to a call center to take place only when the attempt to obtain a subscriber identifier has not been successful or when the user, on his own initiative, directly enters a command for the passing on to take place. In this way, voice dialing can be combined with a conventional directory enquiries service in a premium service.

In a preferred variant of the invention, the communications terminal itself is also fitted with speech recognition means. The possibilities that this opens up are in particular the following: Firstly, the speech recognition means in the communications terminal can be used to activate the terminal to make the first connection for voice communication to the voice dialoging system by means of a spoken command; an activating word. One or more subsequent utterances by the user are then fed to the network-based speech recognition device, i. e. the voice dialoging system, and this then obtains the subscriber identifier by the method according to the invention and sends it back to the user's communications terminal.

In a further variant, the terminal may also have a local voice dialoging system of its own and in this case may execute an advance determination procedure to obtain the desired subscriber identifier. What this means is that the utterances by the user are first processed locally. Only when the advance determination procedure has been unsuccessful and has been brought to an end, for example because the subscriber identifier was not found and/or speech signals input by the user have failed to be recognized, or, where provision is made for this eventuality, there has been a partial failure of this kind, the first connection for voice communication can be made automatically to the voice dialoging system connected to the communications network. What this means is that only when the local processing has failed to produce a result can the relevant utterances be processed for a second time on the network by more versatile or powerful speech recognition means or, if a search in a local telephone directory has been unsuccessful, recourse can be had to the more comprehensive telephone directories of the voice dialoging system or to the databases available on the

network. Once success is achieved, a subscriber identifier is sent back to the communications terminal as in the previous case.

Even though the method according to the invention affords a relatively large number of advantages, it is extraordinarily easy to implement. Apart from a conventional voice dialoging system on the network, which also has to be capable of transmitting the subscriber identifier in machine-readable form, all that need to be present are certain facilities in the communications terminal itself. In programmable terminals, these prerequisites can be produced by suitable software. Where there is a programmable terminal, it is in particular even possible, by altering the software, for the terminal to be retrofitted with a function enabling the method according to the invention to be carried out. In future terminals, the function may even already be implemented in the firmware.

One relatively simple way in which a terminal of this kind can be produced is for a multimode browser to be implemented in the programmable terminal, which may for example be a PC with a telephony function, a PDA or a mobile telephone. In this case, an obvious and especially preferable candidate for use is a multimode browser to the SALT (Speech Application Language Text) specification. In this case, the <salt: listen> tag is implemented in such a way that the spoken language is transmitted to the network-based speaker recognizer, whereas the"call control object"or the corresponding <smex> messages control the local dialing functions of the terminal.

These and other aspects of the invention are apparent from and will be elucidated with reference to the embodiments described hereinafter.

In the drawings: Fig. 1 is a general diagram of a system according to the invention after the making of the first connection for voice communication, Fig. 2 is a general diagram of the system shown in Fig. 1 when the first connection for voice communication has been terminated and the second connection for voice communication has been made, Fig. 3 is a representation of the sequence of events.

As shown in Fig. 1 and Fig. 3, by making a call R, for this purpose, a terminal 2 first makes, via a network N, a connection V, to a server 11 for a voice dialoging system 10. The terminal 2 is a mobile telephone 2 in this case. Any desired other communications

terminals may be used as alternatives, with possible connections to a hard-wired telephone 1 and to a multimode terminal 3 being indicated by way of example by dashed lines. The communications terminal may in particular even be a combination of a mobile telephone and a hands-free facility in an automobile.

The network N may be any desired combination of telephone networks, the internet, GSM, GPRS and UMTS networks, and other mobile telephone networks (though without being limited to a combination of this kind).

The network server 11 is fitted with suitable software 7,8, 9 to enable it to perform its function as a voice dialoging system 10. What are shown here simply by way of indication are a speech recognition module 8, a speech output module (prompt generator) 9 and a control module 7 that operates the speech recognition module 8 and the speech output module 9. The server 11 is also configured to have a list 12 of the possible names, the rules for how they are pronounced and, for each name, an associated subscriber identifier, such as a telephone number or URL for example. It is clear that the server 11 will also be fitted with all the usual hardware and software components that are normally required to operate a server 11 of this kind that is used as a voice dialoging system 10. These include in particular an operating system and further interfaces, to outside databases for example. It is explicitly pointed out that a voice dialoging system 10 of this kind need not necessarily be implemented on a single network server 11 but may also be distributed over a network, in which case various computers may perform widely differing tasks for the voice dialoging system as a whole.

The further course of the method is shown in Fig. 3. While the first speech connection Vl is maintained, one or more utterances by the user are transmitted via the terminal 1 to the voice dialoging system 10. These utterances are recognized by the speech recognition module 8 and are fed to the control module 7, which controls the dialog and, via the prompt generator 9, asks the user to say other things at the appropriate points in time.

What this means is that the control unit 7 controls the entire dialog and ensures that the associated subscriber identifier K is obtained from the list 12 or a database by reference to the information received from the user, which may be the name of a subscriber for example.

This subscriber identifier K is then first encoded (not shown) into a DMTF code and sent to the user's mobile telephone 2.

The identifier K that is sent back is then accepted by a control module 13 in the mobile telephone 2. This control module 13 then ensures that in the next step a call R2 is made to that terminal 5 which is identified by the relevant subscriber identifier K. As a result, a connection V2 is made via the network N to the relevant destination subscriber terminal 5.

This is also shown in Fig. 2. The destination subscriber terminal 4,5, 6 too may be any desired communications terminal. It may, however, also be a further dialoging system or similar device or a further network.

The initialization of the entire process can be performed with a special codeword spoken by the user, which is recognized by a rudimentary speech recognizer (not shown) in the mobile telephone 2 and is processed accordingly in the unit 2. Initialization by the pressing of a key on the keypad of the mobile telephone 2 is equally possible.