|Publication number||USRE42868 E1|
|Application number||US 09/930,395|
|Publication date||Oct 25, 2011|
|Filing date||Oct 25, 1995|
|Priority date||Oct 25, 1994|
|Also published as||CA2202663A1, CA2202663C, CA2372671A1, CA2372671C, CA2372676A1, CA2372676C, CN1164292A, CN1249667C, DE69525178D1, DE69525178T2, DE69535797D1, EP0800698A2, EP0800698B1, EP1172994A2, EP1172994A3, EP1172994B1, US5940793, WO1996013030A2, WO1996013030A3|
|Publication number||09930395, 930395, PCT/1996/2524, PCT/GB/1996/002524, PCT/GB/1996/02524, PCT/GB/96/002524, PCT/GB/96/02524, PCT/GB1996/002524, PCT/GB1996/02524, PCT/GB1996002524, PCT/GB199602524, PCT/GB96/002524, PCT/GB96/02524, PCT/GB96002524, PCT/GB9602524, US RE42868 E1, US RE42868E1, US-E1-RE42868, USRE42868 E1, USRE42868E1|
|Inventors||David J. Attwater, Steven J. Whittaker, Francis J. Scahill, Alison D. Simons|
|Original Assignee||Cisco Technology, Inc.|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (22), Non-Patent Citations (3), Referenced by (2), Classifications (36), Legal Events (1)|
|External Links: USPTO, USPTO Assignment, Espacenet|
1. Field of the Invention
The present invention is concerned with automated voice-interactive services employing speech recognition, particularly, though not exclusively, for use over a telephone network.
2. Related Art
A typical application is an enquiry service where a user is asked a number of questions in order to elicit replies which, after recognition by a speech recogniser, permit access to one or more desired entries in an information bank. An example of this is a directory enquiry system in which a user, requiring the telephone number of a telephone subscriber, is asked to give the town name and road name of the subscriber's address, and the subscriber's surname.
According to one aspect of the present invention there is provided a speech recognition apparatus comprising a store of data containing entries to be identified and information defining for each entry a connection with a word of a first set of words and a connection with a word of a second set of words; speech recognition means; and control means operable:
Preferably the speech recognition means is operable upon receipt of the first voice signal to generate for each identified word a measure of similarity with the first voice signal, and the control means is operable to generate for each word of the list a measure obtained from the measure(s) for the relevant word(s) of the first set (i.e those identified words of the first set with which a word of the list has a common entry). The speech recognition means is then operable upon receipt of the second voice signal to perform the identification of one or more words of the list in accordance with a recognition process weighted in dependence on the measures generated for the words of the list.
The apparatus may also include a store containing recognition data for all words of the second set and the control means is operable following the compilation of the list and before recognition of the word(s) of the list to mark in the recognition data store those items of data therein which correspond to the words not in the list or those which correspond to words which are in the list, whereby the recognition means may ignore all words so marked or, respectively, not marked.
Alternatively the recognition data may be generated dynamically either before recognition or during recognition, the control means being operable following the compilation of the list to generate recognition data for each word of the list. Methods for dynamically generating recognition data fall outside the scope of the present invention but will be clear to those skilled in this art.
Preferably the control means is operable to select for output that entry or entries defined as connected both with an identified word(s) of the first set and an identified word of the second set.
The store of data may also contain information defining for each entry a connection with a word of a third set of words, the control means being operable:
Furthermore, means may be included to store at least one of the received voice signals, the apparatus being arranged to perform an additional recognition process in which the control means is operable:
Preferably the apparatus includes means to recognise a failure condition and to initiate the said additional recognition process only in the event of such failure being recognised.
The apparatus may comprise a telephone line connection; a speech recogniser for recognising spoken words received via the telephone line connection, by reference to recognition data representing a set of possible utterances; and means responsive to receipt via the telephone line connection of signals indicating the origin or destination of a telephone call to access stored information identifying a subset of the set of utterances and to restrict the recogniser operation to that subset.
According to a further aspect of the invention, a telephone apparatus comprises a telephone line connection; a speech recogniser for determining or verifying the identity of the speaker of spoken words received via the telephone line connection, by reference to recognition data corresponding to a set of possible speakers; and means responsive to receipt via the telephone line connection of signals indicating the origin or destination of a telephone call to access stored information identifying a subset of the set of speakers and to restrict the recogniser operation to that subset.
According to a yet further aspect of the invention, a telephone information apparatus comprises a telephone line connection; a speech recogniser for recognising spoken words received via the telephone line connection, by reference to one of a plurality of stored sets of recognition data; and means responsive to receipt via the telephone line connection of signals indicating the origin or destination of a telephone call to access stored information identifying one of the sets of recognition data and to supply this set to the recogniser.
The stored sets may, for example, correspond to different languages or regional accents or, say, two of the sets may correspond to the characteristics of different types of telephone apparatus, for instance the characteristics of a mobile telephone channel.
According to a further aspect of the invention a recognition apparatus comprises
The patterns may represent speech and the recognition means be a speech recogniser.
In accordance with the invention, a speech recognition apparatus comprises
Preferably the first set of signals are voice signals representing spelled versions of the words of the second set or initial portions thereof and the identifying means are formed by the speech recognition means operating by reference to stored recognition information for the said spelled voice signals. Alternatively the first set of signals may be signals consisting of tones and the identifying means is a tone recogniser. The first set of signals may indicate the origin or destination of the receive signal.
In accordance with a further aspect of the invention, a method of identifying entries in a store of data by reference to stored information defining connections between entries and words, comprises
In a further aspect of the invention a speech recognition apparatus comprises
A method of speech recognition by reference to a stored set of words to be recognised, according to the invention comprises
The second signal may also be a speech signal, and the second signal may be recognised by reference to recognition data representing the letters of the alphabet, either individually or as sequences. Alternatively the second signal may be a signal consisting of tones generated by a keypad.
According to another aspect of the invention, a method of speech recognition comprises
Some embodiments of the invention will now be described, by way of example, with reference to the accompanying drawings, in which:
The embodiment of the invention now to be described addresses the same directory enquiry task as was discussed in the introduction. It operates by firstly asking an enquirer for a town name and, using a speech recogniser, identifies as “possible candidates” two or more possible town names. It then asks the enquirer for a road name and recognition of the reply to this question then proceeds by reference to stored data pertaining to all road names which exist in any of the candidate towns. Similarly, the surname is asked for, and a recognition stage then employs recognition data for all candidate road names in candidate towns. The number of candidates retained at each stage can be fixed, or (preferably) all candidates meeting a defined acceptance criterion—e.g. having a recognition score above a defined threshold—may be retained.
Before describing the process in more detail, the architecture of a directory enquiry system will be described with reference to
A main directory database 9 contains, for each telephone subscriber in the area covered by the directory enquiry service, an entry containing the name, address and telephone number of that subscriber, in text form. The town name recognition data store 6 contains, in text form, the names of all the towns included in the directory database 9, along with stored data to enable the speech recogniser 5 to recognise those town names in the speech signal received from the telephone line interface 2. In principle, any type of speech recogniser may be used, but for the purposes of the present description it is assumed that the recogniser 5 operates by recognising distinct phonemes in the input speech, which are decoded by reference to stored data in the store 6 representing a decoding tree structure constructed in advance from phonetic translations of the town names stored in the store 6, decoded by means of a Viterbi algorithm. The stores 7, 8 for road name recognition data and surname recognition data are organised in the same manner. Although, for example, the surname recognition data store 8 contains data for all the surnames included in the directory database 9, it is configurable by the control unit 4 to limit the recognition process to only a subset of the names, typically by flagging the relevant parts of the recognition data so that the “recognition tree” is restricted to recognising only those names within a desired subset of the names.
This enables the ‘recognition tree’ to be built before the call commences and then manipulated during the call. By restricting the active subset of the tree, computational resources can be concentrated on those words which are most likely to be spoken. This reduces the chances that an error will occur in the recognition process, in those cases where one of these most likely words has been spoken.
Each entry in the town data store 6 contains, as mentioned above, text corresponding to each of the town names appearing in the database 9, to act as a label to link the entry in the store 6 to entries in the database 9 (though other kinds of label may be used if preferred). If desired, the store 6 may contain an entry for every town name that the user might use to refer to geographical locations covered by the database, whether or not all these names are actually present in the database. Noting that some town names are not unique (there are four towns in the UK called Southend), and that some town names carry the same significance (e.g. Hammersmith, which is a district of London, means the same as London as far as entries in that district are concerned), an equivalence data store 39 is also provided, containing such equivalents, which can be consulted following each recognition of a town name, to return additional possibilities to the set of town names considered to be recognised. For example if “Hammersmith” is recognised, London is added to the set; if “Southend” is recognised, then Southend-on-Sea, Southend (Campbeltown), Southend (Swansea) and Southend (Reading) are added.
The equivalence data store 39 could, if desired, contain similar information for roads and surnames, or first names if these are used; for example Dave and David are considered to represent the same name.
As an alternative to this structure, the vocabulary equivalence data store 39 may act as a translation between labels used in the name stores 6, 7, 8 and the labels used in the database (whether or not the labels are names in text form).
The use of text to define the basic vocabulary of the speech recogniser requires that the recogniser can relate one or more textual labels to a given pronunciation. That is to say in the case of a ‘recognition tree’, each leaf in the tree may have one or more textual labels attached to it. If the restriction of the desired vocabulary of a recogniser is also defined as a textual list, then the recogniser should preferably return only textual labels in that list, not labels associated with a pronunciation associated with a label in the list that are not themselves in the list.
The system operation is illustrated by means of the flowchart set out in
The next stage is that a further response, relating to the road name, is received (17) from the caller and is processed by the recogniser 3 utilising the data store 7; suppose that five road names meet the recognition criterion. The control unit 4 then instructs the playing (19) of a further message asking for the name of the desired telephone subscriber and meanwhile (20) retrieves from the database 9 a list of the surnames of all subscribers residing in roads having any of the five road names in any of the four geographical locations (and any equivalents), and updating the surname recognition data store 8 in a similar manner as described above for the road name recognition data store. Once the user's response is received (22) by the recogniser, the surname may be recognised (23) by reference to the data in the surname recognition data store.
It may of course be that more than one surname meets the recognition criterion; in any event, the database 9 may contain more than one entry for the same name in the same road in the same town. Therefore at step 24 the number of directory entries which have one of the recognised surnames and one of the recognised road names and one of the recognised town names is tested. If the number is manageable, for example if it is three or fewer, the control means instructs (25) the speech synthesiser to play an announcement from the message data store 3, followed by recitation of the name, address and telephone number of each entry, generated by the speech synthesiser 1 using text-to-speech synthesis, and the process is complete (26). If, on the other hand, the number of entries is excessive then further steps 27, to be discussed further below, will be necessary in order to meet the caller's enquiry.
It will be seen that the process described will have a lower failure rate than a system which chooses only a single candidate town, road or surname at each stage of the recognition process, since by retaining second and further choice candidates the possibility of error due to misrecognition is reduced though there is increased risk of recognition error due to the larger vocabulary. A penalty for this increased reliability is of course increased computation time, but by ensuring that the road name and surname recognition processes are conducted over only a limited number of the total number of road names and surnames in the database, the computation can be kept to manageable proportions.
Moreover, compared with a system in which a second-stage recognition is unconstrained by the results of a previous recognition (e.g. one where the ‘road’ recognition processes is not limited to roads in town proposed system would, when using recognisers (such as those using Hidden Markov Models) which internally “prune” intermediate results, be less liable to prune out the desired candidate in favour of other candidate roads from unwanted towns.
It will be seen too, that the number of possible lists will, in most applications, be so large as to prohibit their preparation in advance, and hence the construction of the list is performed as required. Where the recogniser is of the type (e.g. recognisers using Hidden Markov models) which require setting up for a particular vocabulary, there are two options for updating the relevant store to limit the recogniser's operation to words in the list. One is to start with a fully set-up recogniser, and disable all the words not in the list; the other is to clear the relevant recognition data store and set it up afresh (either completely, or by adding words to a permanent basic set). It should be noted that some recognisers do not store recognition data for ail all words which may be recognised. These recognisers generally have a store of textual information relating to the words that may be recognised but do not prestore data to enable the speech recogniser to recognise words in a received signal. In such so-called “dynamic recognisers” the recognition data is generated either immediately before or during recognition.
The first option requires large data stores but is relatively inexpensive computationally for any list size. The second option is generally computationally expensive for large lists but requires much smaller data stores and is useful when there are frequent data changes. Generally the first option would be preferred, with the second option being invoked in the case of a short list, or where the data change frequently.
The criterion for limiting the number of recognition ‘hits’ at steps 13, 18 or 23 may be that all candidates are retained which meet some similarity criterion, though other criteria such as retaining always a fixed number of candidates may be chosen if preferred. It may be, in the earlier recognition stages, that the computational load and effect on recognition performances of retaining a large town (say) with a low score is not considered to be justified, whereas retaining a smaller town with the same score might be. In this case the scores of a recognised word may be weighted by factors dependent on the number of entries referencing that word, in order to achieve such differential selection.
In the examples discussed above, a list of words (such as road names) to be recognised is generated based on the results of an earlier recognition of a word (the town name). However it is not necessary that the unit in the earlier recognition step or in the list be single words; they could equally well be sequences of words. One possibility is a sequence of the names of the letters of the alphabet, for example a list of words for a town name recognition step may be prepared from an earlier recognition of the answer to the question “please spell the first four letters of the town name.” If recording facilities are provided (as discussed further below) it is not essential that the order of recognition be the same as the order of receipt of the replies (it being more natural to ask for the spoken word first, followed by the spelled version, though it is preferred to process them in the opposite sequence).
It is assumed in the above description that the recognisers always produce a result—i.e. that the town (etc) name or names which give the nearest match(es) to the received response are deemed to have been recognised. It would of course be possible to permit output of a “fail” message in the event that a reasonably accurate match was not found. In this case further action may be desired. This could simply be switching the call to a manual operator. Alternatively further information may be processed automatically as shown in
It is not necessary that the response to be recognised be discrete responses to discrete questions. They could be words extracted by a recogniser from a continuous sentence, for systems which work in this way.
Another situation in which it may be desired to vary the scope of the speech recogniser's search is where it can be modified on the basis not of previous recogniser results but of some external information relevant to the enquiry. In a directory enquiry system this may be a signal indicating the origin of a telephone call, such as the calling line identity (CLI) or a signal identifying the originating exchange. In a simple implementation this may be used to restrict town name recognition to those town names located in the same or an adjacent exchange area to that of the caller. In a more sophisticated system this identification of the calling line or exchange may be used to access stored information compiled to indicate the enquiry patterns of the subscriber in question or of subscribers in that area (as the case may be).
For example, a sample of directory enquiries in a particular area might show that 40% of such calls were for numbers in the same exchange area and 20% for immediately adjacent areas. Separate statistical patterns might be compiled for business or residential lines, or for different times of day, or other observed trends such as global usage statistics of a service that are not related to the nature or location of the originating line.
The effect of this approach can be to improve the system reliability for common enquiries at the expense of uncommon ones. Such a system thus aims to automate the most common or straightforward enquiries, with other calls being dealt with in an alternative manner, for example being routed to a human operator.
As an example,
An extension of this approach is to improve the system reliability and speed for common enquiries, whilst using additional information to enable the less common enquiries to succeed. Thus the less common enquiries are still able to succeed but require more effort and information to be supplied by the caller than the common enquiries require.
As an example consider
If the word is recognised as being outside of the vocabulary or of poor confidence then an additional message 49 is played to ask the caller for more information, which in this case is the first four letters of the town name. Simultaneously, an additional re-recognition of the spoken town name 53 may be performed which can recognise any of the possible town names in the directory. In this example we assume that four town names are recognised 54. At the same time, the caller may be spelling in the first four letters of the town name 50 and two spellings 51 have been confidently recognised. These two spellings are then expanded to the full town names which match them 52. It may be necessary to anticipate common spelling errors, additional or missing letters, abbreviations, and punctuation in the preparation of the spelling vocabulary, and the subsequent matching of the spelt recognition results to the full town names. Assume in this example that five town names match the two spellings.
A comparison 55 identical in purpose to that described in
In other cases, the spoken recognition 53 will be in error and no common words will be found. Alternatively, the recognition of the town name 53, and its subsequent comparison 55, may be considered optional and omitted. In both of these instances the spoken town store will be updated 57 with the five towns derived from the two spellings 52 and the spoken town name re-recognised again 58. In the example, it is assumed that a single confident town name was recognised. This town name may be used to configure the road name recognition data store 7 and the enquiry proceeds as shown in
The deliberate restriction of a vocabulary to only the very most likely words as described above need not necessarily depend on CLI. The preparation of the road name vocabulary based on the recognised town names is itself an example of this, and the approach of asking for additional information, as shown in
Additionally, the re-recognition of a gathered word that has been constrained by additional information such as the four letter spelling in
In the above described embodiment, no account is taken of the relative probability of recognition, for example if the town recognition step 13 recognises town names Norwich and Harwich, then when, at road recognition step 18, the recogniser has to evaluate the possibility that the caller said “Wright Street” (which we suppose to be in Norwich) or “Rye Street” (in Harwich), no account is taken of the fact that the spoken town bore a closer resemblance to “Norwich” than it did to “Harwich”. If desired however, the recogniser may be arranged to produce (in known manner) figures or “scores” indicating the relative similarity of each of the candidates identified by the recogniser to the original utterance and hence the supposed probability of it being the correct one. These scores may then be retained whilst a search is made in the directory database to derive a list of the vocabulary items of the next desired vocabulary that are related to the recognised words. These new vocabulary items may then be given the scores that the corresponding matching word attained. In the case where a word came from a match with more than one recognised word of the previous vocabulary, the maximum score of the two may be selected for example. These scores may then be fed as a priori probabilities to the next recognition stage to bias the selection. This may be implemented in the process depicted in
Step 13. The recogniser produces for each town, a score—e.g.
Similar modification would of course occur for the steps 20, 21, 23. This is just one example of a scheme for score propagation.
The possibility of switching to a manual operator in the event of a “failure” condition has already been mentioned. Alternatively a user could simply be asked to repeat the action that has not been recognised. However, further automated steps may be taken under failure conditions.
A failure condition can be identified by noting low recogniser output “scores”, or of excessive numbers of recognised words all having similar scores (whether by reference to local scores or to weighted scores) or by comparing the scores with those produced by a recogniser comparing the speech to out-of-vocabulary models. Such a failure condition may arise in an unconstrained search like that of the town name recognition of step 13 in
Thus one possible approach is to make provision for recording the caller's responses, and in the event of failure, reprocessing them using the steps set out in
Another possibility is to perform an additional search omitting one stage, and comparing the results as for the ‘spelled input’ case.
If desired, processing using two (or more) such sequences could be performed routinely (rather than only under failure conditions); to reduce delays an additional sequence might commence before completion of the first; for example (in
The technique of storing an utterance and using it in a restricted-vocabulary recognition process following recognition of a later utterance has been described as an option to be used alongside sequential processing, as a cross-check or to provide additional recognition results to be used in the case of difficulty. However, it may be used alone, for example in circumstances where one chooses to have the questions asked in a sequence which seem natural to the user, so as to improve speed and reliability of response, but to process the answers in a sequence which is more suited to the nature of the data. For example in
The use of CLI to modify the expectations of a speech service need not be restricted to the modification of expected vocabulary items as already described. Enquiry systems that require a certain level of security or personal identification may also use CLI to their advantage. The origin of the telephone call as given by the CLI may be used to extract from a store the identity of a number of individuals known to the system to be related to this origin. This store may also contain representative speech which is already verified to have come from these individuals. If there is only one individual authorised to access the given service from the designated origin, or the caller has made a specific claim to identity by means of additional information (e.g. a DTMF or spoken personal identification number) then a spoken utterance may be gathered from the caller and compared with the stored speech patterns associated with that claimed identity in order to verify that the person is who they say that they are. Alternatively, if there are a number of individuals associated with the call origin, the identity of the caller may be determined by gathering a spoken utterance from the caller and comparing it with stored speech patterns for each of the individuals in turn, selecting the most likely candidate that matches with a certain degree of confidence.
The CLI may also be used to access a store relating speech recognition models to the origin of the call. These speech models may then be loaded into the stores used by the speech recogniser. Thus, a call originating from a cellular telephone, for example, may be dealt with using speech recognition models trained using cellular speech data. A similar benefit may be derived for regional accents or different languages in a speech recognition system.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US4701879||Jun 24, 1985||Oct 20, 1987||Standard Telephones And Cables Public Limited Co.||Associative memory systems|
|US4763278||Apr 13, 1983||Aug 9, 1988||Texas Instruments Incorporated||Speaker-independent word recognizer|
|US4947438 *||Jul 11, 1988||Aug 7, 1990||U.S. Philips Corporation||Process for the recognition of a continuous flow of spoken words|
|US5202952 *||Jun 22, 1990||Apr 13, 1993||Dragon Systems, Inc.||Large-vocabulary continuous speech prefiltering and processing system|
|US5267304||Apr 5, 1991||Nov 30, 1993||At&T Bell Laboratories||Directory assistance system|
|US5355474||Sep 27, 1991||Oct 11, 1994||Thuraisngham Bhavani M||System for multilevel secure database management using a knowledge base with release-based and other security constraints for query, response and update modification|
|US5475792||Feb 24, 1994||Dec 12, 1995||International Business Machines Corporation||Telephony channel simulator for speech recognition application|
|US5479488 *||Feb 8, 1994||Dec 26, 1995||Bell Canada||Method and apparatus for automation of directory assistance using speech recognition|
|US5488652 *||Apr 14, 1994||Jan 30, 1996||Northern Telecom Limited||Method and apparatus for training speech recognition algorithms for directory assistance applications|
|US6018736 *||Nov 20, 1996||Jan 25, 2000||Phonetic Systems Ltd.||Word-containing database accessing system for responding to ambiguous queries, including a dictionary of database words, a dictionary searcher and a database searcher|
|CA2091658A1||Mar 15, 1993||Sep 16, 1994||Matthew Lennig||Method and apparatus for automation of directory assistance using speech recognition|
|EP0269233A1 *||Oct 9, 1987||Jun 1, 1988||Smiths Industries Public Limited Company||Speech recognition apparatus and methods|
|EP0299572A2 *||Jul 8, 1988||Jan 18, 1989||Philips Patentverwaltung GmbH||Method for connected word recognition|
|EP0477688A2 *||Sep 12, 1991||Apr 1, 1992||Texas Instruments Incorporated||Voice recognition telephone dialing|
|EP0484070A2||Oct 25, 1991||May 6, 1992||International Business Machines Corporation||Editing compressed voice information|
|EP0533338A2||Aug 6, 1992||Mar 24, 1993||AT&T Corp.||Interface method and apparatus for information services|
|EP0601710A2||Nov 4, 1993||Jun 15, 1994||AT&T Corp.||On demand language interpretation in a telecommunications system|
|EP0625758A1 *||Jan 31, 1994||Nov 23, 1994||International Business Machines Corporation||Natural language processing system|
|GB2165969A *||Title not available|
|JPH06204952A||Title not available|
|WO1993005605A1||Sep 11, 1992||Mar 18, 1993||Bell Atlantic Network Services||Method and system for home incarceration|
|WO1996013030A2 *||Oct 25, 1995||May 2, 1996||British Telecomm||Voice-operated services|
|1||K.E. Niebuhr et al., "N Ary Join for Processing Query by Example Nov. 1976", IBM Technical Disclosure Bulletin, vol. 19, No. 6, Nov. 1976, pp. 2377-2381, XP002081147 New York, US.|
|2||*||Yamada et al., "A Spoken Dialogue System with Active/Non-Active Word Control for CD-ROM Information Retrieval", Speech Communication, 15 (1994) 355-365.|
|3||*||Young, "Use of Dialogue, Pragmatics and Semantics to Enhance Speech Recognition", 8308 Speech Communication 9(1990) Dec., Nos. 5/6 Amsterdam, Netherlands, pp. 551-564.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US8738377 *||Jun 7, 2010||May 27, 2014||Google Inc.||Predicting and learning carrier phrases for speech input|
|US20110301955 *||Dec 8, 2011||Google Inc.||Predicting and Learning Carrier Phrases for Speech Input|
|U.S. Classification||704/231, 704/270.1, 704/E15.044, 704/E15.045, 704/E15.041, 704/275, 704/E17.003, 704/251|
|International Classification||G10L15/22, G10L15/26, H04M3/493, G10L15/10, H04M3/42, G10L15/06, G10L15/18, G10L15/24, G10L17/00|
|Cooperative Classification||G10L15/18, G10L17/00, H04M3/4936, G10L2015/0631, H04M3/42093, G10L15/24, H04M3/42102, H04M2242/22, H04M3/4931, H04M3/42059, H04M2203/355, H04M2201/40, G10L15/26|
|European Classification||G10L15/18, H04M3/493S, H04M3/493D, G10L17/00U, G10L15/24, G10L15/26A|
|Jul 25, 2006||AS||Assignment|
Owner name: BT RAVENSCOURT LLC, VIRGINIA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:BRITISH TELECOMMUNICATIONS PUBLIC LIMITED COMPANY;REEL/FRAME:017982/0951
Effective date: 20041222
Owner name: CISCO RAVENSCOURT L.L.C., DELAWARE
Free format text: CHANGE OF NAME;ASSIGNOR:BT RAVENSCOURT L.L.C.;REEL/FRAME:017982/0967
Effective date: 20050321
Owner name: CISCO TECHNOLOGY, INC., CALIFORNIA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:CISCO RAVENSCOURT LLC;REEL/FRAME:017982/0976
Effective date: 20060710