GB2293519A - Voice processing system - Google Patents

Abstract

A voice processing system 60 is connected via a digital trunk line 100 to a telephone switch 10, which is itself connected to a telephony network 30 by a digital trunk line 20. The digital trunk line between the switch and the network uses a first encoding law for representing the amplitude of the telephony signals, for example A-law encoding. However, applications running on the voice processing system are designed to operate on signals encoded in accordance with a different amplitude-encoding law, for example mu-law encoding. The voice processing system therefore includes in a telephone interface module 70 conversion means for converting incoming signals from the first encoding law to the second, and vice versa for outgoing signals. <IMAGE>

Description

VOICE PROCESSING SYSTEM The present invention relates to a voice processing system that in operation is connected to a telephone switch by a digital trunk line.

Modern telephony signals are generally transmitted in digital form, using standard 8-bit samples at a rate of 8kHz, thereby requiring an overall bandwidth of 64 kbits per second. The nature of audio signals is such that an essentially logarithmic quantisation allows the best representation of the original analog input. Two quantisation laws in particular are widely used; an A-law compression (or companding) in Europe, and a mu-law compression in North America, and are defined by recommendation G.711 in the CCITT "Yellow Books".

To allow large volumes of telephone traffic to be handled simply, individual signals are time multiplexed together for transmission over trunk lines. In North America, the standard form of trunk line is known as T1, and provides 24 simultaneous lines, each capable of handling a single 64 kbits per second telephone signal. In Europe, the standard form of trunk line is known as El, and provides 30 simultaneous lines. Several T1 or El lines can be grouped together in a trunk line to provide greater overall capacity.

It can be seen that North America uses T1 trunks with mu-law companding, whilst Europe uses El trunks with A-law companding (as specified in CCITT recommendations G.733 and G.732 respectively). To allow transatlantic telephone calls, it is necessary to be able to convert (transcode) effectively between these two systems. Suitable conversion schemes are well-known in the art, see for example US 3825924, US 4404544, US 4595907, and US 4661946.

The present invention is particularly concerned with voice processing systems. These devices, which are well-known in the art (see for example "Voice Processing", by Walt Teschner, published by Artech House), perform a variety of functions, the most common of which is voice mail (also known as voice messaging), whereby callers who cannot reach their intended target can instead record a message for them for subsequent retrieval. Another important form of voice processing system is a voice response system (VRU), in which callers interact with the VRU, generally by means of pressing DTMF keys, in order to obtain desired information.Other features that have been or are being incorporated into voice processing systems include voice recognition (typically so that callers can enter responses into the system without having to use DTMF keys), and text to speech, whereby ASCII data can be read out to a caller. It should be appreciated that there are no clear dividing lines between the different categories of voice mail system, VRU, and so on; rather the term voice processing system is used herein to broadly denote any type of system which can sit at or terminate one end of a telephone line. Generally voice processing systems are implemented on general purpose computers, with additional telephone interface hardware. Often an application running on the computer is specially developed by the customer (ie the owner of the voice processing system) to control the interaction between the caller and the voice processing system.This is particularly true for VRUs which normally need to be customised to meet the requirements of any given customer. Frequently, the development of the controlling application program represents a significant investment of time and money by the customer.

The simplest voice processing systems have as their input a conventional analog telephone line, in other words, they can plug into a socket in place of a normal telephone set. Although effective in some situations, such systems have limited call capacity (ie the number of incoming telephone lines which can be simultaneously supported is relatively low), and are also restricted in the type of service that they can operate.

The more sophisticated voice processing systems have a digital trunk connection to a switch. Such voice processing systems are generally installed at the sites of customers who have a relatively large volume of incoming or outgoing telephone calls and therefore have their own switch.

Thus the voice processing system makes or receives telephone calls through the switch over one or more El or T1 lines as appropriate.

The trunk lines not only carry the actual audio telephone conversations, but can also be used to provide a limited degree of signalling between the switch and the voice processing system. This is referred to as channel associated signalling (CAS). For example, the sort of information that can be conveyed between the switch and the voice processing system includes the telephone number of the call being made, or notification of the arrival of a call. The protocols used in this signalling have unfortunately not been standardised, but rather vary according to the country and switch involved. This creates a problem for the manufacturers of voice processing systems, who must therefore try to support a wide range of protocols.

One particular difficulty that can arise occurs when a customer wishes to take an application developed in one country and to use it in a second country. As explained above, the ability to re-use an application in another country is very attractive to a customer, since it saves the time and cost of having to develop another version of the application.

Moreover, it is often desirable for the customer to use the same switch in all countries, which may well allow them to negotiate a lower price for the switches, and is also likely to reduce service costs for the customer. An additional benefit is that by maintaining the same combination of application program and switch in all countries, compatability problems are minimised (eg an application may exploit a particular feature of the chosen switch).

Thus it is highly attactive for a multinational customer to be able utilise the same pairing of switch and application throughout their organisation. However, this can lead to problems for the company that supplies the voice processing systems, and in particular if the voice processing system supports a signalling protocol used by the switch in a first country, but does not support the signalling protocol used by the same switch in another country. In such circumstances, the provider of the voice processing system may be faced with having to develop support for a new signalling protocol - a costly and time-consuming business.

Accordingly, the invention provides a voice processing system for attachment to a switch by a digital trunk line, whereby telephony signals encoded using a first encoding law are exchanged between the switch and the voice processing system over said digital trunk line, and characterised in that said voice processing system is operable to support at least one process which generates output or requires input in the form of a telephony signal having a second encoding law, and said voice processing system includes means for converting telephony signals between the first and second encoding laws.

The provision of a voice processing system which includes the ability to convert between a first encoding law and a second encoding law offers a broad range of benefits. It allows a voice processing system designed for operation under one encoding law to be used largely unmodified with another encoding law thus for example said at least one process may comprise voice recognition software which can now be ported from one country to another without having to worry about the encoding law being used. Depending on the countries involved, the first encoding law is typically A law and the second encoding law is mu law, or vice versa.

A voice processing system in accordance with the invention also helps to resolve the above-mentioned signalling problem, in that the type of trunk connection between the switch and the voice processing system can be varied, thereby offering a greater range of available protocols.

Thus it is possible for the trunk line between the switch and the voice processing system to be of a form normally matched with the second encoding law rather than said first encoding law, eg providing A law encoding on a T1 trunk line.

In a preferred embodiment, the voice processing system comprises a telephone interface module for attachment to said trunk line, and the conversion means is incorporated with the telephone interface module.

Performing the conversion at a relatively low level in the voice processing system is computationally efficient, and allows the majority of the voice processing system to utilise the same telephony signals encoded using the second encoding law.

A possible alternative to this is where the voice processing system comprises a telephone interface module for attachment to said trunk line, and a device driver for exchanging data between the telephone interface module and the rest of the voice processing system, and the conversion means is incorporated into said device driver. Again it is relatively simple to implement the conversion in the device driver, although this can have some drawbacks in terms of performance.

It should be appreciated that the invention is independent of the precise architecture of the switch or network. Thus it is applicable to voice processing systems attached to conventional switches, ATM switches and so on. Likewise, it is not limited to configurations in which channel associated signalling is used, but can also be employed in common channel signalling environments, such as ISDN.

An embodiment of the invention will now be described in detail by way of example with reference to the following drawings: Figure 1 is a simple block diagram showing a voice processing system connected to a telephone switch; Figure 2 illustrates the main software components of the voice processing system of Figure 1; Figure 3 is a very schematic diagram illustrating the operation of the switch; Figure 4 is a more detailed diagram of the structure of the voice processing system of Figure 1; and Figures 5A and 5B illustrate the flow of information through and out of the telephone interface module for incoming and outgoing data respectively.

Figure 1 is a simple block diagram showing a switch 10 which exchanges telephony signals with the external telephone network 30 over digital trunk line 20. Attached to the switch are a plurality of conventional telephone extensions 40; these are of no direct relevance to the present invention and so will not be described further. Also attached to the switch via a digital trunk line 100 is a voice processing system 60. In the current implementation, the voice processing system is a DirectTalk/6000 system (ie runs the DirectTalk/6000 software), but the same principles apply whatever voice processing system is being used.

The DirectTalk/6000 system comprises two main hardware components, a telephone interface module 70 (also termed a digital trunk processor), and computer workstation 80, which in the case of the DirectTalk/6000 system is a RISC System/6000 workstation. Also shown is an adapter card 90, which provides an interface between the RISC System/6000 workstation and the telephone interface module. Note that in many voice processing systems, the telephone interface module is incorporated into the adapter card for attachment direct to the computer workstation. The DirectTalk/6000 system (software plus hardware) is available from IBM Corporation, and is described more fully in IBM Callpath DirectTalk/6000 General Information and Planning (reference number GC22-0100-03), also available from IBM.As stated above, although the invention is being described with reference to the DirectTalk system, it is applicable to many other voice processing systems.

Figure 2 is a simple block diagram of the main software components of a DirectTalk/6000 system. Running on the RISC System/6000 is first of all the operating system 110 for the workstation, which in the present case is AIX, and then the DirectTalk/6000 software 120 itself. Finally, also running on the RISC System/6000 workstation is an application 130, generally developed by the customer, which interacts with the operating system and the DirectTalk/6000 software to provide the desired telephony function. Various routines 140 also run within the digital trunk processor 70. These routines are downloaded from the RISC System/6000 onto the telephone interface module when the digital trunk processor is enabled, and handle items such as detection of tones, silence, and voice, and the generation of tones.There is also some software on the adapter card 90, which primarily serves simply to transfer data between the RISC System/6000 workstation and the telephone interface module. This software is of no direct relevance to the present invention and so will not be discussed further.

In the current implementation, the switch 10 is an Aspect switch.

The configuration shown in Figure 1, with an Aspect switch and a DirectTalk/6000 voice processing system has been successfully installed in the United States, where trunk lines 20 and 100 are standard T1 lines.

It was now desired to install the same system in Europe, whereby trunk lines 20 and 100 would be El lines. However, this led to a problem in that the Aspect switch could only support one signalling protocol for El lines, a protocol known as DAS, which is defined by British Telecom, whereas the DirectTalk/6000 system does not support this protocol.

One way around this problem of course would be to add support for an extra protocol to either the switch or the voice processing system, so that there was a common signalling protocol that they could both use. The drawback with this approach however, is that it can be expensive and time-consuming to develop such extra signalling support, and moreover additional homologation may be required (homologation is the process whereby manufacturers of telephony equipment obtain permission to connect their equipment to the telephone networks of different countries).

Another approach would be to replace either the Aspect switch of the DirectTalk/6000 system with units that did provide the correct match of signalling protocols. However, as explained above this solution is unattractive, since it reduces the ability of the customer to obtain an advantageous price on the switch and voice processing systems by buying in bulk, and moreover complicates worldwide support and service due to the heterogeneous nature of systems being used. A further consideration is that the application software 130 was specially designed for the combination of the Aspect switch and DirectTalk/6000 hardware, and porting it to another platform would again consume time and money.

One possible solution which was investigated was to interpose a protocol converter on trunk line 100 between the switch and the voice processing system. This would then convert DAS signals received from the switch into a signalling protocol that the DirectTalk/6000 voice processing system could understand, and vice versa. Although in theory this would allow correct signalling between the switch and the voice processing system, in practice it was found that the converter had a serious impact on the performance of the overall installation, and so did not offer a viable solution.

A brute force solution would be to put a El to T1 converter in front of the switch (ie on trunk line 20). However, such conversion equipment is expensive and would appear to be very wasteful, since it would also perform conversion on telephone signals to and from the telephone extensions 40 when there is no need for this. Note that it would not be possible to locate such conversion equipment between the switch and the voice processing system, since this would not solve the problem about incompatible signalling protocols.

The approach adopted by the present invention is to modify the link 100 between the switch and the voice processing system so that it is a T1 line (as opposed to the El trunk 20). As shown schematically in Figure 3, the design of the switch 10 is such that this is relatively easy to perform. The switch essentially functions as TDM bus. Thus for incoming calls to the DirectTalk system, samples 320 arriving from the network are read off from the various incoming channels on El trunk lines, with each channel being assigned its own slot 310 on the TDM bus. The samples are then read out from the slots and directed to the appropriate channel on the T1 trunk. Since a switch is in fact primarily designed for channel routing, this sort of operation is straightforward on most switches.

Although the signals are now grouped together in T1 trunks of 24 channels each, the encoding law used to represent the actual voice samples is still A-law, as received from the network. There is no facility in the switch to convert these signals to mu-law (some switches can perform the complete conversion between Tl/mu law and E1/A law, but the complexity of doing this adds considerably to the cost of the switch). Thus the DirectTalk/6000 system ends up receiving (and having to output) signals encoded in A-law but grouped into T1 trunk lines.

Figure 4 is a schematic diagram of the main components of a DirectTalk/6000 system. Only those components relevant to an understanding of the present invention will be described; further details can be found in the above-mentioned manual, plus the other manuals referenced therein. The first set of components run on the RISC System/6000 workstation 80. These comprise a device driver 580 which is used to interact via the adapter card (not shown) with the telephone interface module 70, and a system configuration module 510, which is used to specify configuration information, including whether a T1 or El connection is being used. Other processes in the DirectTalk/6000 system examine this configuration information and deduce from the T1 or El connection the type of encoding is being used.Thus the type of encoding is closely coupled in the system to the type of trunk connection.

A state table process 520 effectively provides the program control of applications executing in the DirectTalk/6000 system (ie in developing an application, the customer creates a set of state tables). The channel processor 530 contains the code which performs the actions specified by the state tables 520. A custom server 550 unit allows external connections into and out of the DirectTalk/6Od0 system. It is shown as split into two portions, because generally the customer has to write some code to interact with it. Typically the custom server is used to provide an interface to a voice recognition or text to speech system (TTS) 560.

Further included on the RISC System/6000 workstation are a system monitor process 540, which provides run-time supervisory and administrative functions, and a set of tools 570; since these are of no direct relevance to the present invention they will not be described further.

Figure 4 also depicts the contents of the telephone interface module in more detail. In particular, this module comprises two cards, a trunk interface card (TIC) 450 and a V-Pack card 460. There is one V-Pack and one TIC for each T1/E1 trunk connection between the switch and the voice processing system. The TIC includes an ACFA chip 430, which is an industry standard chip available from Siemens, and is used to detect or insert signalling for incoming or outgoing calls respectively on trunk line 100. The V-pack contains a set of six digital signal processors (DSPs), comprising one master DSP 410 and five associated slave DSPs 420.

Each slave DSP processes six channels within a T1/E1 trunk, so that the set of five slave DSPs can handle all thirty channels for an El trunk (in the case of a T1 trunk connection one of the slave DSPs is not required).

The slave DSPs operate on data having a linear amplitude scale, ie with no companding, as will be explained in more detail below.

The data flow between the master and slave DSPs, and also the device driver 580, is shown in more detail in Figures 5A and 5B for incoming and outgoing data respectively. Considering Figure 5A first, incoming data first arrives at the master DSP, from where it is routed to either the slave DSP selected for that channel, or to the device driver for handling by the RISC System/6000. In the former case, since the slave DSP requires linear data, the master DSP effectively decodes (unpacks) the companded signal to its linear form. This is performed using a table 610 lookup; standard tables are available for doing this, either from mu law or A law, and the result is a 12 bit signal representing the linear amplitude.The consequence of this is that the routines executing on the slave DSP, which perform functions such as voice activity detection and tone detection, are independent of whether the incoming signal is A-law or mu-law encoded, since such they operate only on decoded linear data.

The slave DSP can also perform compression on the linear input voice data, which can then be passed to the device driver for storage (eg if the DirectTalk system was being used to provide voice mail).

If no compression is being used however, data is transmitted straight from the master DSP through to the device driver. The master DSP executes a routine to convert or transcode the data from 8-bit A law to 8-bit mu law. Again, this can be performed using a look-up table 620, and so is very fast. From then on, the remainder of the DirectTalk/6000 system can process the incoming signal as if it had been received from a conventional T1 trunk using mu law encoding.

The output procedure is slightly different in that all data from the device driver is routed first via the master DSP to the appropriate slave DSP, and then onto the master DSP. The reason for doing this is that the slave DSP provides automatic volume adjustment on the outgoing signal, to ensure that the amplitude of the outgoing signal is within accepted limits for the telephone network. Outgoing data are first expanded from 8-bit mu-law to a linear twelve bit signal by the master DSP, again using a look-up table 630. Thus the signal received by the slave DSP is again linear. Note that if the signal being output was in fact compressed, then it would be transmitted directly from the master to the slave without the need for any decoding, allowing the slave to perform the decompression.After processing by the slave, the linear signal is then returned to the master, which encodes it back down from 12 bit linear to 8 bit A law for transmission out on the trunk line to the switch. Thus for outgoing data the master DSP simply uses decoding and encoding procedures that are standard for T1 and El trunks respectively and so the routines already exist in the DirectTalk/6000 voice processing system; the difference in the present arrangement is that the decoding up to 12 bits is performed according to a different law from the encoding back down to 8 bits, thereby producing as a net effect the desired transcoding operation.

Returning now to the original concern, namely the selection of a suitable signalling protocol between switch and the voice processing system, the chosen protocol is known as the Release Link Tie protocol (this is defined by the switch manufacturer Aspect). This protocol is essentially designed for T1 trunks, and is supported by both the DirectTalk/6000 system and the Aspect switch; note that in general some form of approval may be required from the telecommunications companies in order to adopt such a configuration, although full homologation may not be necessary.

The direct conversion between A-law and mu-law (and vice versa) performed using a look-up table introduces some distortion to the telephone signal. This distortion, which is best regarded as quantisation noise, in fact turns out to be roughly the same as the distortion introduced by the original encoding from linear into A-law or mu-law.

Since telephone signals are nearly always generated or received in a noisy environment anyway, the slight extra distortion introduced by the conversion process of the present invention is effectively negligible.

Note also that it is only the first transcoding the leads to any increase in distortion; subsequent transcodings do not introduce any further distortion.

The above-described arrangement has proved to be a very effective way of allowing a T1/A law connection to be supported by a voice processing system. Essentially the only changes needed to the DirectTalk/6000 system are as follows: (i) The system configuration process now allows the independent specification of both the trunk line type (El or T1) and the external encoding format (A law and mu law).However, the majority of the DirectTalk system does not know about, and so does not use this extra information; in other words, most processes continue to imply the type of encoding from looking at the trunk type in the configuration data; (ii) The device driver is slightly altered, such that when the V-Pack is enabled, for example by the system monitor under user control, the configuration data is examined to determine the type of encoding; dependent on this, a file is then referenced which contains the names and locations of the appropriate modules to download onto the V-pack, plus the transcoding tables where necessary; in addition, the device driver also sets one particular bit in the master DSP memory which is used to store an indication of whether or not transcoding is employed; (iii) A new A-law to Mu law conversion routine has been written for the master DSP, which performs the appropriate conversion on incoming data as it is transmitted from the master DSP to the device driver; the master DSP examines the above-referenced bit in memory to determine whether or not to switch on this routine.

Thus this approach requires very little modification to the DirectTalk/6000 voice processing system, whilst providing very satisfactory performance. However, other arrangements might also be implemented; for example it is quite possible for the device driver to perform the conversion instead of the master DSP. This is satisfactory in principle, but was found in practice to adversely affect overall system performance.

It should be appreciated that although the development of the above conversion process was motivated by the need to provide a suitable form of signalling between the switch and the voice processing system, there are many other situations in which the invention will be beneficial. For example, an organisation may have developed some voice recognition software in one country, using one encoding law, which they then want to utilise in a second country using a different encoding law. The invention permits them to achieve this without modification to the application. In such a situation there will typically be no need for a hybrid trunk line between the switch and the voice processing system; for example, a conventional El trunk with A-law encoding could be used providing suitable signalling could be provided, but with the conversion occurring in the voice processing system to let the application use mu-law encoding. Note that the invention may also be used to allow pre-recorded digital voice segments to be re-used in the second country; thus currently the DirectTalk system stores all such pre-recorded segments in compressed form since this avoids the need to worry about which encoding law to use. In accordance with the present invention, this need no longer be a concern, although of course language differences also impose practical restrictions on the re-use of voice segments across national boundaries.

Claims (8)

1. A voice processing system (60) for attachment to a switch (10) by a digital trunk line (100), whereby telephony signals encoded using a first encoding law are exchanged between the switch and the voice processing system over said digital trunk line, and characterised in that said voice processing system is operable to support at least one process which generates output or requires input in the form of a telephony signal having a second encoding law, and said voice processing system includes means (70) for converting telephony signals between the first and second encoding laws.

2. The voice processing system of claim 1, wherein the voice processing system comprises a telephone interface module (70) for attachment to said trunk line, and the conversion means is incorporated with the telephone interface module.

3. The voice processing system of claim 1, wherein the voice processing system comprises a telephone interface module (70) for attachment to said trunk line, and a device driver (580) for exchanging data between the telephone interface module and the rest of the voice processing system, and the conversion means is incorporated into said device driver.

4. The voice processing system of any preceding claim, wherein the means for converting includes a look-up table (630).

5. The voice processing system of any preceding claim, wherein said at least one process which generates output or requires input in the form of a telephony signal having a second encoding law is a voice recognition or text to speech process (560).

6. The voice processing system of any preceding claim, wherein the first encoding law is A-law and the second encoding law is mu-law.

7. The voice processing system of any of claims 1 to 5, wherein the first encoding law is mu-law and the second encoding law is A-law.

8. The voice processing system of any of preceding claim, wherein the trunk line between the switch and the voice processing system is of a form normally matched with the second encoding law rather than said first encoding law.