GB1581477A

GB1581477A - Apparatus for synthesising verbal announcements

Info

Publication number: GB1581477A
Application number: GB1964077A
Authority: GB
Original assignee: Post Office
Current assignee: Post Office
Priority date: 1978-05-19
Filing date: 1978-05-19
Publication date: 1980-12-17

Description

(54) APPARATUS FOR SYNTHESISING VERBAL ANNOUNCEMENTS (71) We, THE POST OFFICE, a British Corporation established by Statute of 23 Howland Street, London W1P 6HQ, do hereby declare the invention, for which we pray that a patent may be granted to us, and the method by which it is to be performed to be particularly described in and by the following statement: This invention relates to apparatus for synthesising verbal announcements.

In telephone exchanges, computer man/ machine interfaces, and other similar environments it is often required to transmit to subscribers or users pre-recorded announcements.

The number of different announcements required may be numerous, and this is particularly true of announcements which contain a word or phase which is varied according to circumstances. For example, the announcement 'the dialling code is ' has a constant part and a variable part, and to prerecord separately a complete range of these announcements differing only in the variable part would necessitate large storage facilities.

It has been recognised that storage requirements for both variable and constant announcements could be greatly reduced if the announcements were to be synthesised on demand from a basic component store of some kind. However, in practice, both constructional difficulties and the subjective unacceptability of the synthesised speech have hitherto proved difficult to overcome.

Accordingly, in a first aspect the present invention provides apparatus for synthesising verbal announcements including: a vocabulary store for storing, in digitally encoded form, information representative of speech, the information being stored as a number of information units each unit representing respectively a speech segment of the same duration; a conduction path along which digital signals representative of the contents of the vocabulary store are cyclically broadcast, the whole contents of the vocabulary store being broadcast in a period equal to the duration of one of said speech segments; and means for selecting sequences of information units from the conduction path to form selected encoded announcements.

The information stored in the vocabulary store may be digitally encoded pre-recorded speech.

The means for selecting sequences of information segments may include an announcement assembly store for storing the order of words required to synthesise preselected verbal announcements.

The apparatus may include a further store which is adapted to temporarily contain specifications of a word or part of words in terms of the component information units representative of the word or part in the vocabulary store, the further store being operative to temporarily store words or parts or words that are to be used as variable elements in an announcement being delivered.

The means for selecting sequences of information units from the conduction path includes one or more output control processors, the or each processor being associated with one or more output channels and being controllably operative to select sequences of information units which form selected announcements and allow them to be transmitted from said conduction path to its associated output channel or channels, the or each processor having access to the announcement assembly store to obtain the information enabling the process to select the sequences of speech segments appropriate to selected announcements.

The apparatus may form part of a telephone exchange wherein each of a plurality of said output control processors controls a plurality of output channels, which output channels are connectable via decoding means to telephone lines so that synthesised announcements may be selectively transmitted to receiving units in the telephone network.

Embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings. In the drawings: Figure 1 is a block schematic diagram of a vocabulary store and output channels in an apparatus for synthesising verbal annoucements; Figure 2 is a diagrammatic representation of the method of transmitting the contents of the vocabulary store of Figure 1; Figure 3 is a block schematic diagram of an output channel arrangement of the apparatus; Figure 4 is a block schematic diagram showing a number of the output channels of Figure 3 and their connections with the vocabulary store and other parts of the apparatus, and Figure 5 is a diagrammatic representation of the operational relationship between announcement assembly tables of the apparatus and the vocabulary store.

Apparatus for synthesising verbal announcements in a telephone exchange includes a vocabulary store 1 which feeds via a speech highway 2 a number of output channels. The output channels synthesise particular announcements from units of speech encoded as signals on the speech highway 2 and supplied by the speech store 1.

The vocabulary store 1 contains a large number of storage loops la which may be in the form of charge-coupled devices, or alternatively, MOS random access or magnetic bubble memories. Each loop contains, in digitally encoded form, a unit of speech. These units of speech, stored as digitally encoded information units, are of equal duration and will be termed 'speech segments'. Since these segments do not necessarily correspond to phonemes (the natural speech units) the intelligibility of an announcement synthesised from them imposes a limit on their duration. It has been found that the duration of a speech segment should not exceed 1/8 second if objectional effects are to be avoided in the synthesised speech.

It has also been found that speech synthesised from digitally encoded units has an acceptable quality to the human ear when the digital information is provided at a rate of 32 x 1024 bits/sec or more. This figures is however dependent on the quality of speech from which the digitally encoded speech is derived. It has been found that by improving this original quality an information rate as low as 24 x 1024 bits/sec can be used. In describing the present embodiment it will however be assumed that an information rate of 32 x 1024 bits/sec. is the most appropriate.Thus, if speech segments of 1/8 second are used these may suitably be represented by information units containing 4 x 1024 bits of informaion. An information unit that represents identical words from different announcements is not duplicated in the speech store, thereby reducing the storage capacity required, although if the sound of a word or syllable differs significantly according to the context of its use then different versions of it are stored.

The storage loops in the vocabulary store 1 cyclically feed their information in synchronism to the highway 2. Each cyclically repeated output from the speech store 1 contains all the speech segments represented within the storage loops. This output is shown diagrammatically in Figure 2, where at the start of a cycle to the first bits of each segment are fed in sequence, then the second bits, and so on. In this way the contents of the store are supplied to the highway in 1/8 second cycles giving an information rate along the highway of S 32 x 1024 bits/sec, where S is the number of information units in the vocabulary store. To achieve the maximum storage capacity, with a finite highway capacity, the segment duration should be as large as is allowable by the subjective criteria.The duration of the segments however also affects the usefulness of a particular segment, i.e., if a segment is too long it is likely to be appropriate for one announcement only, thereby tending to increase the amount of speech required to be stored.

The speech highway 2 comprises a number of twisted pairs, and for a store of 1024 segments, i.e., 4Mbits, 8 such twisted pairs are used each carrying serially interleaved 128 speech information units at an information rate of 4096 x 1024 bits/sec. Thus a combination of space and time division multiplexing is used to transmit the contents of the vocabulary store along the speech highway.

Referring now to Figure 3, each output channel is connected to speech highway 2 by a space switch 4 and a time switch 5, these select respectively a particular twisted pair of the highway and a particular speech information unit on that twisted pair. The space and time switches 4 and 5 are controlled by signals from a segment address register 6 which is in turn controlled by a channel group microprocessor 7 via its highway 8 and a channel control port 9.

The segment address register 6 is also connected to a highway 10 along which clocking signals having the segment duration and in synchronism with the segments are transmitted.

A channel decoder 11 connects the switches 4 and 5 to the output channel proper 12. The group control microprocessor 7 transmits signals over the highway 8 to control the channel control port 9. These signals include, a space switch address, a time switch address, and clocking signals to load these addresses into the channel control port 9. At the beginning of each segment cycle to a clock pulse with a rising edge corresponding to to is generated by the clock (Figure 4) and is transmitted over the highway 10 to load the space and time switch addresses into the segment address register 6.

The correct address of the information unit representing the appropriate segment to be delivered on a particular channel is selected by the microprocessor 7 by means of an announcement synthesis algorithm stored as a program in the microprocessor.

Each group control microprocessor 7 controls 128 output channels, as is shown in Figure 4.

Referring to Figure 4, a random access announcement assembly table 12 contains stored information specifying the sequence of segments required from the vocabulary store 1 to reconstitute particular announcements.

The announcement assembly table is con nected to each of the channel group control microprocessors 7. Each of the microprocessors 7, in addition to storing the announcement synthesis algorithm, maintains a record for each of the channels under its control. This record includes: (i) the channels status, i.e. whether or not it is currently delivering an announcement; (ii) the identity of the announcement required on that channel, if active; (iii) the variable data required by that channel, e.g. a telephone number; (iv) the identity of the segment currently being delivered by that channel; (v) the identity of the word currently being delivered by that channel; (vi) the number of times the announcement is to be repeated; and (vii) temporary storage space.

The microprocessors 7 receive instructional information from a central processor in the telephone exchange via an interface unit 15.

These instructions are either to connect or to disconnect a particular channel to or from an announcement. A 'correct' message from the exchange includes information specifying: (i) the identity of the channel on which an announcement is to appear; (ii) the identity of the announcement required; (iii) the variable data required for that announcement, if any; and (iv) the number of times an announcement is to be repeated.

If variable data is required on a channel the group control microprocessor 7 allocates space in a variable data table 14 to that channel and loads pointers to this stored variable data into the channel's record. The operational relationship between the announcement assembly table 13, the variable data table 14 and the vocabulary store is illustrated in Figure 5.

Each microprocessor 7 executes the announcement synthesis algorithm once every segment cycle for each of its active channels.

With the information specifying the sequence of words in an announcement from the announcement assembly table 13, and the channel records, the microprocessors execute the algorithm to load the channel control port 9 of each active channel with the address of the next information unit to be delivered. If variable data is to be delivered, the announcement assembly table 13 specifies the point at which this is required and the variable data is obtained from its table 14 in the sequence defined by the pointers in the channel record.

The microprocessor then executes the algorithm for each word stored in the variable data table and generates the appropriate series of information unit addresses. At the end of each segment cycle the address of the next information unit is loaded into the corresponding channel segment address register 6.

Before the apparatus can be operated the vocabulary store 1, and the announcement assembly table 13 need to be loaded with information. The first step in this loading process is the tape-recording of a basic speech vocabulary. Each required word is recorded on a magnetic tape and separated by a length of blank tape. The contents of this tape are communicated to an input terminal 1 6a of a store loading interface unit 16 where they are digitally encoded by a delta modulator within this unit. The interface unit 16 also incorporates a digital circuit for detecting the blank spaces between the recorded words on the magnetic tape, thereby providing an indication of the word boundaries. The digitally encoded words are transmitted to the speech store 1 in a known sequence and the successive segments of the words are stored on loops having consecutive addresses.One of the group control microprocessors 7a records the number of segments used to store each of the words.

In alternative embodiments the basic speech vocabulary may be digitally encoded separately from the apparatus and then stored on magnetic tape. The vocabulary store may then be loaded directly from this tape.

The announcement assembly table 13 is loaded with information specifying the order of words required for particular announcements.

To carry out this loading the required announcements are translated, by an operator, into a coded form in which each word in the announce ment is represented by its position in the sequence of words supplied to the vocabulary store 1. These coded announcements are recorded on a paper tape. The contents of the paper tape are fed to the input terminal 1 6b of the interface unit 16 and then to the microprocessor 7a which translates the coded words into their respective segment addresses. Since each word in the speech store vocabulary is stored as a set of successive segments, to identify a word the microprocessor 7a need only identify the address of the first segment of the word and the number of segments which constitute the word.The microprocessor 7a thus loads the announcement assembly table 13 with the required announcements in a predetermined sequence, with each announcement being coded as a sequence of words, and each word identified by its starting segment and its length in segments.

The variable data table, when loaded with data from the control interface 15 via a group control microprocessor, similarly stores the required words in terms of their starting segment addresses and their durations.

It will be appreciated that in the above described embodiment announcements transmitted by the output channels will always being at the correct starting point of the announcement and may be repeated by any predetermined number of times. It will also be appreciated that the apparatus has the capacity to be extended to operate additional output channels and that this extension may be accomplished very simply.

WHAT WE CLAIM rS:- 1. Apparatus for synthesising verbal announcements including: a vocabulary store for storing, in digitally encoded form, information representative of speech, the information being stored as a number of information units each unit representing respectively a speech segment of the same duration; a conduction path along which digital signals representative of the contents of the vocabulary store are cyclically broadcast, the whole contents of the vocabulary store being broadcast in a period equal to the duration of one of said speech segments; and means for selecting sequences of information units from the conduction path to form selected encoded announcements.

2. Apparatus for synthesising verbal announcements as claimed in claim 1 wherein the information stored in the vocabulary store is digitally encoded pre-recorded speech.

3. Apparatus for synthesising verbal announcements as claimed in claim 2 wherein the vocabulary store is loaded from a magnetic tape recording of the words required to be stored, the words being recorded in a known sequence and being digitally encoded prior to their transmission to the vocabulary store.

4. Apparatus for synthesising verbal announcements as claimed in any one of the preceding claims wherein the means for selecting sequences of information units includes an announcement assembly store for storing the order of words required to synthesise preselected verbal announcements.

5. Apparatus for synthesising verbal announcements as claimed in claim 4 including a further store which is adapted to temporarily contain specifications of a word or part of words in terms of the component information units representative of the word or part in the vocabulary store, the further store being operative to temporarily store words or parts of words that are to be used as variable elements in an announcement being delivered.

6. Apparatus for synthesising verbal announcements as claimed in any one of the preceding claims wherein each word stored in the vocabulary store comprises a set of segments that are represented by a set of said information units that are in a successive relationship in the cyclical broadcast of the contents of the store.

7. Apparatus for synthesising verbal announcements as claimed in claim 6 when dependent on claim 4 wherein the words forming parts of announcements stored in the announcement assembly store are defined in that store in terms of a) the position of the information unit representative of the first segment of the word in the cyclical sequence on which the information units are broadcast and b) the number of segments that the word comprises.

8. Apparatus for synthesising verbal announcements as claimed in claim 6 wherein the means for selecting sequences of information units from the conduction path includes one or more output control processors, the or each processor being associated with one or more output channels and being controllably operative to select sequences of information units which form selected announcements and allow them to be transmitted from said conduction path to its associated output channel or channels, the or each processor having access to the announcement assembly store to obtain the information enabling the processor to select the sequences of speech segments appropriate to selected announcements.

9. Apparatus for synthesising verbal announcements as claimed in any one of the preceding claims wherein the digital signals representative of the contents of the vocabulary store are broadcast cyclically along the conduction path using time or space division multiplexing or a combination of both.

10. Apparatus for synthesising verbal announcements as claimed in any one of the preceding claims wherein the vocabulary store comprises a number of charge-coupled devices, each charge-coupled device storing one or more encoded speech segments.

11. Apparatus for synthesising announcements as claimed in claim 4 or claim 8 wherein the announcement assembly store is loaded by a processor that is, in use, connected to receive information specifying the order of words required to produce a number of announcements, and being operative to load this information into the announcement assembly store.

12. Apparatus for synthesising verbal announcements as claimed in claim 11 when dependent on claim 8 wherein said processor which loads the announcement assembly store is one of said output control processors.

13. Apparatus for synthesising verbal announcements as claimed in claim 11 or claim 12 each when dependent on claim 2 wherein said information specifying, to the processor which loads the announcement assembly store, the order or words required to produce a number of announcements is coded so that words are specified in terms of their respective position in the sequence of words supplied to the vocabulary store, these spec ifications being translated by the processor to a form in which the words are specified in terms of the address of an information unit representative of at least one of their component segments in the vocabulary store, the information loaded into the announcement assembly store being of this form.

**WARNING** end of DESC field may overlap start of CLMS **.

Claims

**WARNING** start of CLMS field may overlap end of DESC **. be appreciated that the apparatus has the capacity to be extended to operate additional output channels and that this extension may be accomplished very simply. WHAT WE CLAIM rS:-

1. Apparatus for synthesising verbal announcements including: a vocabulary store for storing, in digitally encoded form, information representative of speech, the information being stored as a number of information units each unit representing respectively a speech segment of the same duration; a conduction path along which digital signals representative of the contents of the vocabulary store are cyclically broadcast, the whole contents of the vocabulary store being broadcast in a period equal to the duration of one of said speech segments; and means for selecting sequences of information units from the conduction path to form selected encoded announcements.

14. A telephone exchange including

apparatus for synthesising verbal announcements as claimed in claim 8 wherein each of a plurality of said output control processors controls a plurality of output channels, which output channels are connectable via decoding means to telephone lines so that synthesised announcements may be selectively transmitted to receiving units in the telephone network.

15. Apparatus for synthesising verbal announcements substantially as hereinbefore described with reference to the accompanying drawings.