Patent Application
20050131676
1. Statement of the Technical Field
The present invention relates to the field of computer software and speech recognition and more particularly to user-navigated dynamic voice portals that use speech recognition technology.
2. Description of The Related Art
Contrary to visual applications, voice-based applications have the problem that for input recognition no strict pattern matching can be used. The nature of speech recognition makes it very difficult to distinguish between terms having similar pronunciations. Therefore, during the design of speech applications, care should be taken to provide input choices which are pronounced as differently as possible, so as to avoid the problem of recognizing the wrong choice.
The problem of recognizing the wrong input choice in a speech recognition application occurs with voice portals, which are generally built by various parties that may not be aware of the terms used in the various applications disposed within the voice portal. Often, a voice portal will have, in addition to the current grammars (or commands) for the actual choice to be made, additional active grammars, such as certain “universal” grammars that allow a user to navigate through the portal, e.g. a command such as “go back.” Thus, at any given moment, a combined set of grammars are active, and the voice recognition engine has to search in the set of combined active grammars for a match.
A problem arises if the various grammars used across the various applications on the portal are designed by different parties, as is the case for voice portals built on a general portal architecture, such as the IBM WebSphere™ Portal Server. General portal architecture allows for new applications to be added dynamically by an administrator. The new added choices created by each new application modify the available choices in a selection menu, and thereby affect the quality of recognition. Generally, the administrators are not voice technology specialists, and may further have to operate a voice portal in multiple languages. Because of this, there is always a risk that a new voice application may drastically reduce the quality of the portal.
Below these menu options are the various portlets or voice applications in the applications phase 70 at the bottom of the menu. Applications 20, 22, 24 each branch off from menu item 16, while applications 40, 42, and 44 each branch off from menu item 18. The two sets of voice applications may have been written and arranged by different parties not knowing which terms the other party used for the title of each application. Within each branch of applications additional grammars would be added to the active set which the speech recognition engine of the portal must recognize.
In menu 10, it can be seen that application 34 is titled “Directory,” which is the same as menu option 18. If the grammar for selecting menu option 18 is active within the selection choice following menu option 17, then the system would have trouble distinguishing between identically pronounced terms. Similarly, if a universal grammar such as “store settings” was also active, this would present recognition problems if the user were to navigate through menu item 18, which has the application named “Stores.”
Currently, the only way of testing a portal's recognition quality after setting up the portal or installing a new voice application (or portlet) is to call into the system and check manually, or by user testing with a human user, how well the system works. This can be time-consuming and expensive. It would be desirable therefore, to provide a quality evaluation tool that assesses the ability of a voice portal to recognize different terms in the various applications attached to the portal, by analyzing and measuring the similarity of the terms.
The present invention addresses the deficiencies of the art with respect to evaluating the quality of voice input recognition by a voice portal and provides a novel and non-obvious method, system and apparatus for evaluating the quality of voice recognition by dynamic voice portals.
In a method of evaluating the quality of voice input recognition by a voice portal, a current grammar is extracted from the voice portal. A test input is generated for the current grammar. In this regard, the test input includes a test pattern and a set of active grammars for the current grammar. The test input can be entered into the voice server and the test pattern can be analyzed against the set of active grammars with a speech recognition engine in the voice server. Consequently, a measure of the quality of recognition for the current grammar can be derived.
Systems consistent with the present invention include a system for evaluating the quality of voice input recognition by a voice portal. An analysis interface extracts a set of current grammars from the voice portal. A test pattern generator generates a test input for each current grammar. The test input includes a test pattern and a set of active grammars corresponding to each current grammar. The system further includes a text-to-speech engine for entering each test pattern into the voice portal. A results collector analyzes each test pattern entered into the voice portal with the speech recognition engine against the set of active grammars corresponding to the current grammar for said test pattern. A results analyzer derives a set of statistics of a quality of recognition of each current grammar.
Additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The aspects of the invention will be realized and attained by means of the elements and combinations particularly pointed out in the appended claims. It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.
The accompanying drawings, which are incorporated in and constitute part of the this specification, illustrate embodiments of the invention and together with the description, serve to explain the principles of the invention. The embodiments illustrated herein are presently preferred, it being understood, however, that the invention is not limited to the precise arrangements and instrumentalities shown, wherein:
The present invention is a method and system for evaluating the quality of voice input recognition by a voice portal. The invention works by collecting a set of grammars for one or more voice applications disposed in a voice portal and testing the ability of the voice portal to recognize a particular grammar from among the set of other grammars that may be active with the particular grammar being tested. A measure of quality of recognition can be derived for each grammar, thereby enabling the voice portal to be reconfigured to allow for better voice input recognition.
The portal server 105 can be voice-enabled through coupling to a voice server 110. The voice server 110 is the unit with which an outside caller directly communicates, and can be linked to a telephone network or some other communications network. The voice aggregator 107 is the software that manages the various voice applications 108 running on the portal server 105. When a user communicates with the voice portal, the voice aggregator presents the user with a menu, such as the menu in
The analysis interface 120 exposes external entities to the logic of the voice aggregator 107 and allows the grammar and dependencies collector 140 to collect the various grammars enabled in the voice portal. The grammar database 145 and measurements results database 150 can be one or more data storage media or devices. The signal manipulator 112 can be any signal processing component that emulates the influence of different telephone or communications network qualities, such as line length, crosstalk, or noise, that is applied to the output of the TTS generator 114. The TTS 114 and manipulator 112 can be separated from the voice server 110 or can be integral to the voice server 110.
As used herein, a “current grammar” shall mean any grammar that is on the system of the voice portal, and can be any one of the grammars that corresponds to the various menu options for: (i) navigating through the voice portal, and (ii) selecting one of the portlets 108 on the portal server 105. The core idea of the invention is to check all current grammars in a voice portal with an automatic mechanism, so as to assess the capability and quality of voice recognition of the voice portal.
The test pattern generator 125 software can select a grammar from the set of current grammars stored in database 145, as well as the other grammars dependent on the selected grammar. A dependent grammar is any other grammar that may be executed by a user at any given aggregation step when navigating through the menu of the voice portal. Taking the menu in
For each selected current grammar, the test pattern generator 125 creates a “test input” for the grammar, at step 220. The test input can include both a test “pattern” and a set of active grammars corresponding to the current grammar for which the test input and test pattern is generated. The test pattern can be the actual word or term for the current grammar, or may also include additional words, terms, or sounds. The test pattern can also be entire sentences or phrases. Thus, the test input can include one or more test patterns that incorporate the selected current grammar in some way.
The test pattern generator 125 thus generates a test input for each current grammar and also aggregates a set of active grammars corresponding to the current grammar for each test input. The test input can be a VXML document having the test patterns and set of active grammars incorporated therein.
The test input is then entered into the voice server at step 230. The test pattern itself is entered through the TTS engine 114 and signal processor 112 into the voice server 110. The signal processor 112 can manipulate the sound of the test pattern by emulating the effects of different user voices, different languages, varying communications network qualities, and other modifications of the sound signature of the test pattern. Both TTS engine 114 and signal manipulator 112 may be separate units outside of the voice portal, in which case the synthesized output of the two units could be connected to the voice server 110 through some communications network. Or, the TTS 114 and signal manipulator 112 may already be integrated within the voice server 110. The set of active grammars corresponding to the current grammar for which the test pattern is generated is entered into the voice server 110 through a separate channel, such as from the results collector servlet 130, and may be done through the VXML test document described hereinabove.
Once the test pattern is entered into the voice server 110, in step 240, a speech recognition engine in the voice server can be used to obtain an assessment of how well the voice portal recognized the test pattern. The quality of the recognition of the test and the current grammar being tested by the test input is therefore obtained. This quality of recognition can be monitored and collected by the results collector servlet 130 and stored in the measurements results database 150. The quality of recognition can include a set of statistics that are generally used to assess the quality. Two examples of such statistics are the confidence level and n-best results, which generally used by speech recognition engines. Thus, the set of statistics can include a confidence level and a set of n-best results for the test input for each grammar tested, and resulting the confidence level and set of n-best results for the test input can be compared with an expected value for each metric to assess the quality of recognition.
In step 250, the process determines whether the quality of recognition is acceptable. If the quality is not acceptable, system 100 can be used to adjust and modify the selected current grammar, re-execute the test phase by running through steps 210, 220, 230 and 240, and re-assess whether the quality of recognition is acceptable. If the results are found to be acceptable at step 250, the process terminates.
An example of the process of the method of the present invention can be illustrated using the voice portal menu 10 of
Confidence levels of close to one are regarded as a near perfect match, whereas confidence levels of near zero are regarded as not a match. If more than one grammar in the set of active grammars were to produce very high confidence levels, each above a certain pre-determined threshold, then the quality of recognition could be assessed as poor, since the system could incorrectly recognize one grammar for another. This can be seen in the example set above, where the two grammars for “Directory” each produce confidence levels that are far above any other of the grammars. The voice portal would therefore recognize one of the two grammars having the high confidence level. But it would not be able to distinguish between the two. Thus, the system would show that the quality of recognition is low in that the voice portal would not be able to easily distinguish between two grammars for two different commands. Hence, the user's ability to navigate through the portal would be compromised.
The present invention therefore provides a method and system for evaluating the quality of voice input recognition by a voice portal. The present invention can execute a test of the voice portal very quickly, at relatively low cost, and with far greater ease than a human system administrator of a voice portal could otherwise do. The present invention could test all grammars in a system, even if the grammars were spoken in different languages, and even if a voice portal system administrator does not know the languages. Furthermore, because of the ability of TTS engines to render different voices (male, female, fast, slow . . . ), the present invention can utilize the TTS engine to test the voice portal with a much more robust input, than a human administrator can otherwise do. Also, because of a speech recognition engine's more fine-grained ability to characterize the similarity of two sounds, while a human system administrator could only determine whether a voice portal simply worked or did not work, the present invention can measure how much one sound differs from another to produce a more detailed assessment of the quality of recognition by a voice portal.
The present invention can be realized in hardware, software, or a combination of hardware and software. An implementation of the method and system of the present invention can be realized in a centralized fashion in one computer system, or in a distributed fashion where different elements are spread across several interconnected computer systems. Any kind of computer system, or other apparatus adapted for carrying out the methods described herein, is suited to perform the functions described herein.
A typical combination of hardware and software could be a general purpose computer system with a computer program that, when being loaded and executed, controls the computer system such that it carries out the methods described herein. The present invention can also be embedded in a computer program product, which comprises all the features enabling the implementation of the methods described herein, and which, when loaded in a computer system is able to carry out these methods.
Computer program or application in the present context means any expression, in any language, code or notation, of a set of instructions intended to cause a system having an information processing capability to perform a particular function either directly or after either or both of the following a) conversion to another language, code or notation; b) reproduction in different material form. Significantly, this invention can be embodied in other specific forms without departing from the spirit or essential attributes thereof, and accordingly, reference should be had to the following claims, rather than to the foregoing specification, as indicating the scope of the invention.
