Patent Grant
8954325
The present invention relates to information services, and in particular to improving speech recognition in information services automation systems.
Information services systems have been implemented since the beginning of telephony communications. For various reasons, and historically based on the need for directory assistance, telephony subscribers could call an information services system, request particular information, and receive the information. As communications evolve, the sophistication of the information services systems and the type of information provided with these systems has significantly increased. Currently, information services systems provide all types of information, from traditional directory numbers and addresses to driving directions and movie listings.
As the need for information services increases, information services providers have implemented automated systems that are capable of handling certain requests in a fully automated fashion, without requiring operator assistance, by utilizing technologies such as speech recognition, speech synthesis, recorded speech playback, and digit detection. Naturally, there are numerous reasons, such as varying accents, dialects, and languages, which prevent these automated systems from being able to properly respond to all requests. As such, the requests that are not recognized or otherwise handled properly may be sent to a human operator, who will interact with the caller and provide the requested information.
Given the significant cost savings associated with automation, there is a continuing need to provide more accurate and reliable automation. The primary hurdle in automation is the difficulty in recognizing speech due to the various languages, accents, dialects, and pronunciations of words that formulate the caller's request for information. At this time, the speech recognition engines in these information services automation systems are only updated periodically, and these updates are not necessarily based on actual use, but rather on general predictions involving speech recognition patterns. Further, there is no mechanism to provide feedback to the automation system based on actions taken by the operator after the automation system has failed. There is a need to provide feedback to the automation system based on the operator's interaction with the caller to improve speech recognition, and thus the ability to automate future requests in a more effective manner.
The present invention allows feedback from operator workstations to be used to update databases used for providing automated information services. When an automated process fails, recorded speech of the caller is passed on to the operator for decision making. Based on the selections made by the operator in light of the speech of or other interactions with the caller, a comparison is made between the speech and the selections made by the operator to arrive at information to update the databases in the information services automation system. Thus, when the operator inputs the words corresponding to the speech provided at the information services automation system, the speech may be associated with those words. The association between the speech and the words may be used to update different databases in the information services automation system.
In one embodiment, the automation process involves processing the speech to detect phonemes, using the phonemes to detect words, and using the words to detect an entry that is associated with the information being requested by the caller. If there is a failure at any one of these detection stages, the speech is sent to the operator. When the operator listens to the speech and provides operator input corresponding to the words or entries, the various databases used to look up words based on phonemes, entries based on words, or information based on entries may be updated. As such, a word typed in by the operator may be associated with a group of phonemes for the speech. Similarly, an entry may be associated with a new word or group of words. The information services automation system may send information identifying the step in the automation process where the automation failed. As such, the particular database to update based on the operator input can be selected based on the point of failure.
Those skilled in the art will appreciate the scope of the present invention and realize additional aspects thereof after reading the following detailed description of the preferred embodiments in association with the accompanying drawing figures.
The accompanying drawing figures incorporated in and forming a part of this specification illustrate several aspects of the invention, and together with the description serve to explain the principles of the invention.
The embodiments set forth below represent the necessary information to enable those skilled in the art to practice the invention and illustrate the best mode of practicing the invention. Upon reading the following description in light of the accompanying drawing figures, those skilled in the art will understand the concepts of the invention and will recognize applications of these concepts not particularly addressed herein. It should be understood that these concepts and applications fall within the scope of the disclosure and the accompanying claims.
Turning now to
During the transition from the information services automation system 16 to the operator workstation 18, the initial audible information provided by the caller that was recorded by the information services automation system 16 will be transferred to the operator workstation 18, such that the operator may listen to the recorded information without having to ask the caller to repeat the information. If necessary, the operator may communicate with the caller to clarify information or obtain additional information to assist in obtaining the requested information.
Once the requested information is obtained by the information services automation system 16 or by an operator at one of the operator workstations 18, a voice session is established between the caller and an automated audio system 20. The automated audio system 20 will then interact with the information services automation system 16 or the operator workstation 18 to obtain the requested information and deliver the requested information to the caller in a synthesized fashion. Notably, the functionality of the automated audio system 20 may be integrated with the information services automation system 16 or the operator workstations 18.
The information services automation system 16, operator workstations 18, and automated audio system 20 may communicate and cooperate with each other via any number of networks or signaling conventions. For the present invention, when the information services automation system 16 fails to provide a fully automated request, the results of the subsequent operator assistance is fed back to the information services automation system 16 to update the various databases used for automation in a manner increasing the likelihood that subsequent information requests will be automated. To initiate information services requests, a caller may use any type of telephony terminal 22 and initiate a voice session, such as a traditional telephone call, to information services wherein the call will be directed to the information services automation system 16 via the information services switch 14.
Turning now to
The original speech is broken into phonemes, which are converted to words by the endpoint detection function 26. The words are then converted to available entries by the recognition detection function 30. Different words or word sequences may be associated with a given entry. For example, the entry of “Joey's Pub and Pizza” may be associated with the following word or words: 1) Joey's Restaurant, 2) Joey's Pub, 3) Joey's Bar, 4) Joey's Pizza, 5) Joey's Pizza Pub, and 6) Joey's Pizza and Pub. The recognition detection function 30 and the grammar database 32 may be configured wherein each of the entries are not necessary but if a certain number of words match a decision is made on a desired entry. Once the entry is determined, the search function 34 will access the information database 36 to obtain the associated information. In this instance, the information may include directory assistance information including the directory number and address for Joey's Pizza and Pub, driving directions, menu information, specials, or any other information that may be desirable to provide to the caller or requested by the caller. Once the requested information is obtained, it is sent to the automated audio system 20 for delivery to the caller.
There are many potential points of failure in the automation process. For example, the speech detection function 24 may not be able to select phonemes, the endpoint detection function 26 may not be able to determine a word or words, the recognition detection function 30 may not be able to detect an entry, and the search function 34 may not be able to determine information for a given entry. If there is a failure at any of these points, a store and forward function 38 will send a recording of the speech to an available operator workstation 18. The store and forward function 38 may also indicate the type of failure or the point of failure in the automation process for the associated speech.
Turning now to
Instead of stopping the automated processing system at this point as with traditional systems, the present invention continues by providing feedback to the information services automation system 16 based on the decisions made by the operator. Accordingly, a comparison is made between the operator input and the provided failure information, which may relate to the phonemes, words, or entries, depending on when the failure occurred (step 110). The failure information may represent the point of failure, or may include the phonemes, words, or entries associated with the automation failure. Thus, the comparison of the failure information to the operator entry can take place on the appropriate level, such as the phoneme level, the word level, or the entry level. The databases are updated (step 114) and the process ends. For example, if the endpoint detection function 26 was unable to detect a word based on the given phonemes, the words associated with the entry ultimately provided by the operator may be associated with the group of phonemes of the request and added to the dictionary database 28. Thus, a new group of phonemes may be associated with an existing word, or a new word may be added to the dictionary database 28 in association with the group of phonemes. In the latter case, the grammar database 32 would be updated with a new word to associate with the entry as well. Once the comparison is made, database information is generated to update the pertinent databases, such as the dictionary database 28, the grammar database 32, and the information database 36 (step 112). In an effort to keep the databases from growing too large, the additional information resulting from feedback from the operator workstations 18 may be removed after a certain period of time or if available memory stores become low.
From the above, the recorded speech that was unable to be processed by the information services automation system 16 is compared in light of an entry provided by the operator at the operator workstation 18. The comparison may be between the given speech and the entry, the recorded speech and the words of the entry, as well as the recorded speech and the phonemes corresponding to the words of the entry. Thus, the actual speech recognition aspect associated with the dictionary database 28 and the grammar database 32 may be updated as well as the basic information database 36. Those skilled in the art will recognize variations in the different information services automation systems 16 and realize different ways to provide feedback for updating the information services automation system 16 in light of the above teachings.
Turning now to
Turning now to
Those skilled in the art will recognize improvements and modifications to the preferred embodiments of the present invention. All such improvements and modifications are considered within the scope of the concepts disclosed herein and the claims that follow.
| Number | Name | Date | Kind |
|---|---|---|---|
| 4979206 | Padden et al. | Dec 1990 | A |
| 5033088 | Shipman | Jul 1991 | A |
| 5163083 | Dowden et al. | Nov 1992 | A |
| 5329608 | Bocchieri et al. | Jul 1994 | A |
| 5535120 | Chong et al. | Jul 1996 | A |
| 5875426 | Bahl et al. | Feb 1999 | A |
| 5897616 | Kanevsky et al. | Apr 1999 | A |
| 5915001 | Uppaluru | Jun 1999 | A |
| 5991364 | McAllister et al. | Nov 1999 | A |
| 6138100 | Dutton et al. | Oct 2000 | A |
| 6182045 | Kredo et al. | Jan 2001 | B1 |
| 6195641 | Loring et al. | Feb 2001 | B1 |
| 6253181 | Junqua | Jun 2001 | B1 |
| 6442519 | Kanevsky et al. | Aug 2002 | B1 |
| 6446076 | Burkey et al. | Sep 2002 | B1 |
| 6470357 | Garcia, Jr. et al. | Oct 2002 | B1 |
| 6477500 | Maes | Nov 2002 | B2 |
| 6480819 | Boman et al. | Nov 2002 | B1 |
| 6487534 | Thelen et al. | Nov 2002 | B1 |
| 6493671 | Ladd et al. | Dec 2002 | B1 |
| 6532446 | King | Mar 2003 | B1 |
| 6553113 | Dhir et al. | Apr 2003 | B1 |
| 6671670 | Levin et al. | Dec 2003 | B2 |
| 6714631 | Martin et al. | Mar 2004 | B1 |
| 6792096 | Martin et al. | Sep 2004 | B2 |
| 6799163 | Nolan | Sep 2004 | B2 |
| 6801893 | Backfried et al. | Oct 2004 | B1 |
| 6873951 | Lin et al. | Mar 2005 | B1 |
| 6882707 | Engelke et al. | Apr 2005 | B2 |
| 6911916 | Wang et al. | Jun 2005 | B1 |
| 6944593 | Kuzunuki et al. | Sep 2005 | B2 |
| 6999923 | Ablondi et al. | Feb 2006 | B1 |
| 7047193 | Bellegarda | May 2006 | B1 |
| 7149694 | Harb et al. | Dec 2006 | B1 |
| 7174300 | Bush | Feb 2007 | B2 |
| 7191133 | Pettay | Mar 2007 | B1 |
| 7197459 | Harinarayan et al. | Mar 2007 | B1 |
| 7243072 | Kotzin | Jul 2007 | B2 |
| 7277851 | Henton | Oct 2007 | B1 |
| 7398209 | Kennewick et al. | Jul 2008 | B2 |
| 7401023 | Schwartz et al. | Jul 2008 | B1 |
| 7437295 | Pitts et al. | Oct 2008 | B2 |
| 7447636 | Schwartz et al. | Nov 2008 | B1 |
| 7487095 | Hill et al. | Feb 2009 | B2 |
| 7542904 | Chestnut et al. | Jun 2009 | B2 |
| 7555431 | Bennett | Jun 2009 | B2 |
| 20020120447 | Charlesworth et al. | Aug 2002 | A1 |
| 20030105638 | Taira | Jun 2003 | A1 |
| 20030130847 | Case | Jul 2003 | A1 |
| 20030233235 | Park | Dec 2003 | A1 |
| 20040049388 | Roth et al. | Mar 2004 | A1 |
| 20040249636 | Applebaum et al. | Dec 2004 | A1 |
| 20040267528 | Roth et al. | Dec 2004 | A9 |
| 20050100153 | Pines et al. | May 2005 | A1 |
| 20050102142 | Soufflet et al. | May 2005 | A1 |
| 20050203738 | Hwang | Sep 2005 | A1 |
| 20050221812 | Gailey et al. | Oct 2005 | A9 |
| 20070043566 | Chestnut et al. | Feb 2007 | A1 |
| 20070233487 | Cohen et al. | Oct 2007 | A1 |
| 20100057457 | Ogata et al. | Mar 2010 | A1 |
| Entry |
|---|
| Non-Final Office Action for U.S. Appl. No. 10/774,083 mailed Jul. 14, 2006, 8 pages. |
| Final Office Action for U.S. Appl. No. 10/774,083 mailed Dec. 29, 2006, 12 pages. |
| Advisory Action for U.S. Appl. No. 10/774,083 mailed Mar. 2, 2007, 3 pages. |
