Patent Application
20060167684
The present invention relates to a speech recognition method and system, and more particularly to a speech recognition method and system in which the recognition results could be confirmed or corrected.
The results of the speech recognition often contain a number of errors. Currently, there are two ways to deal with the errors. One is to re-input the whole speech by the user for correction. The other is to correct the errors with the correcting dialog method specified by the speech recognition system, which requires the user to input the speech one by one for speech recognition and confirmation. Both of the ways are undesirable because the user has to spend lots of time on the confirmation and correction processes.
Please refer to
Generally, the conventional speech recognition method as depicted in
Without display interface, the clues are raised by the system via producing speech for the user. In this way, not only some errors might be caused due to the mis-hearing by the user, but a lot of time is required for the system to raise the clues via speech. If parts of the results are erroneously judged during the speech recognition in the case that more than one value of speech is allowed to be inputted into the system at the same time, the correction can be made either through re-inputting the whole speech by the user or through the correcting dialog method specified by the speech recognition system. Both of the two ways are time-consuming. Besides, the recognition results of the re-inputted speech are not guaranteed to be completely correct.
With display interface, the delay and inaccuracy resulting from the speech interface can be avoided. That is, the recognition results can be shown on the display interface so that the user can judge whether the recognition results are correct or not. However, the correction for the recognition results could only be made by the speech interface. This is completely the same as the speech recognition system without display interface.
Additionally, more and more advanced multimedia data storage/playing devices are available in the market, which are capable of storing lots of data or playing plenty of programs. Therefore, it is more and more difficult to do the search and retrieval for the data or programs.
Presently, the search and retrieval method for data or programs on the portable device is to press the buttons thereon to select the desired function from the menu. This could be achieved by directly pressing the buttons on the portable device or by employing the buttons on the remote controller, e.g. the function control button or the channel selection button for the recorder or television. Owing to the limitation for the number of buttons on the portable device, the display interface with a hierarchical menu is often used for assistance. Such a complicated hierarchical menu not only becomes a nuisance for the user but is inefficient.
There are also more and more intelligent portable devices available in the market. Take the personal digital assistant (PDA) for example, it could record a lot of data, such as telephones and addresses, personal calendars, personal notebooks, MP3 files, radio channels and so on. The functions and commands of the portable device are increasing, but the number of buttons thereon is not correspondingly increased due to the limitation for the volume thereof. Moreover, the display of the portable device is too small to show all of the functions and commands thereon, not to mention the difficulty for the user to memorize so many commands. Hence, it is desirable to employ the speech recognition as the input interface for the portable device.
Even though the employment of the speech recognition as the input interface is more natural for the user, there are still many problems to be solved, however. For example, the recognition results usually contain a number of errors, and the method for correcting these errors is inefficient, which bring the user a serious inconvenience while using the portable device. Therefore, it is of great urgency to develop a better and more convenient speech recognition method and system therefor.
In order to overcome the drawbacks in the prior art, a novel speech recognition method and system are provided. The particular design in the present invention not only solves the problems described above, but also is easy to be implemented.
In accordance with one aspect of the present invention, a speech recognition method and system for the portable device are provided. In the speech recognition system, a displaying device is used for displaying the recognition results, and a locking device is used for confirming the recognition results.
In accordance with another aspect of the present invention, a speech recognition method and system for the portable device are provided. In the speech recognition system, a specific region of the displaying device serves as the communication interface for language understanding, and a keypad is used for confirming/correcting the recognition results.
In accordance with a further aspect of the present invention, a speech input method and system for the portable device are provided. The portable device is capable of being connected to a remote server via the wireless network to access the database of the remote server. In this way, not only the capacity of the database in the portable device can be economized, but the efficiency thereof can be reinforced.
In accordance with further another aspect of the present invention, a speech recognition method is provided. The speech recognition method includes steps of (a) receiving a speech from a user and recognizing the speech for generating a plurality of recognition results, (b) displaying the recognition results for the user to lock correct values in the recognition results, (c) determining whether the correct values are sufficient for searching a database, (d) saving the correct values as known values to narrow the recognition range and repeating step (a) to step (c) when the correct values are insufficient for searching the database, and (e) searching the database for a desired datum based on the correct values when the correct values are sufficient.
Preferably, the recognition results are shown on a displaying device.
Preferably, the displaying device is a touch screen.
Preferably, the correct values in the recognition results are locked by the user with a locking device.
Preferably, the locking device is one selected from a group consisting of a button, the touch screen and a remote controller.
Preferably, the known values are stored in a storage device.
Preferably, the storage device is a register.
Preferably, the database is one selected from a group consisting of a memory, a flash disk, a hard disk and a remote server.
Preferably, the speech recognition method further includes a step of re-recognizing the speech from the user when a part of the correct values is known.
In accordance with further another aspect of the present invention, a speech recognition method is provided. The speech recognition method includes steps of (a) displaying a plurality of fields on a displaying device, wherein each of the field corresponds to an attribute, (b) inputting a speech by a user based on the attribute, (c) recognizing the speech to generate a plurality of recognition results, (d) displaying the recognition results in corresponding fields for the user to lock correct values in the recognition results with a locking device, (e) determining whether the correct values are sufficient for searching a database, (f) saving the correct values as know values to narrow the recognition range and repeating step (b) to step (e) when the correct values are insufficient for searching the database, and (g) searching the database for a desired datum based on the correct values when the correct values are sufficient.
Preferably, the speech recognition method further includes a step of re-recognizing the speech from the user when a part of the correct values is known.
Preferably, the speech recognition method further includes a step of automatically searching for the desired datum without completely filling the fields.
In accordance with further another aspect of the present invention, a speech recognition system is provided. The speech recognition system includes a speech input device for receiving a speech from a user, a speech recognition device connected to the speech input device for recognizing the speech to generate a plurality of recognition results, a displaying device connected to the speech recognition device for displaying the recognition results, a locking device connected to the displaying device for the user to lock correct values in the recognition results, a storage device for saving the correct values as known values, and a database for storing a desired datum to be searched according to the correct values.
Preferably, the displaying device is a touch screen.
Preferably, the locking device is one selected from a group consisting of a button, the touch screen and a remote controller.
Preferably, the storage device is a register.
Preferably, the correct values are saved as the known values via the storage device when the correct values are insufficient.
Preferably, the database is one selected from a group consisting of a memory, a flash disk, a hard disk and a remote server.
Preferably, the desired datum is searched from the database based on the correct values when the correct values are sufficient for searching the database.
In accordance with further another aspect of the present invention, a speech recognition method is provided. The speech recognition method includes steps of (a) receiving a speech from a user and recognizing the speech for generating a plurality of recognition results, (b) displaying one pair of the recognition results for the user to confirm/correct the recognition result, (c) repeating step (b) until all of the recognition results are confirmed/corrected by the user, and (d) searching for a desired datum based on the confirmed/corrected recognition results.
Preferably, the recognition results are shown one by one on a specific region of a displaying device.
Preferably, the recognition results are shown as an ‘attribute-value’ format.
Preferably, the attributes and said values are confirmed/corrected one by one by the user via a control device.
Preferably, the control device is one selected from a group consisting of a keypad, a remote controller and a personal digital assistant.
Preferably, the keypad includes a recording/playing button, an accepting button, a rejecting button, an attribute-correcting button and a value-correcting button.
Preferably, the speech recognition method further includes a step of searching for the desired datum based on the confirmed/corrected attributes and the confirmed/corrected values after one of the attributes and the values is confirmed/corrected.
Preferably, the speech recognition method further includes a step of determining whether the attributes and the values which are not confirmed/corrected need to be confirmed/corrected continuously.
In accordance with further another aspect of the present invention, a speech recognition system is provided. The speech recognition system includes an input device for receiving a speech from a user, a speech recognition understanding device connected to the input device for generating a plurality of recognition results in response to the speech, a confirmation/correction module connected to the speech recognition understanding device for confirming/correcting the recognition results, a displaying device connected to the confirmation/correction module for displaying the recognition results one by one on a specific region thereof, a control device connected to the confirmation/correction module for the user to confirm/correct the recognition results, and a search module connected to the confirmation/correction module for searching for a desired datum based on the confirmed/corrected recognition results.
Preferably, the speech recognition system further includes a storage/receiving device for storing the datum.
Preferably, the datum is one of a digital datum and a video program.
Preferably, the input device is a microphone.
Preferably, the speech recognition understanding device includes a speech recognition device and a language understanding device.
Preferably, the speech recognition device performs a speech recognition based on a lexicon.
Preferably, the language understanding device performs a language understanding based on a grammar rule.
Preferably, the recognition results are shown as an ‘attribute-value’ format.
Preferably, the confirmation/correction module is an interactive meaning confirmation/correction software.
Preferably, the control device is one selected from a group consisting of a keypad, a remote controller and a personal digital assistant.
Preferably, the keypad includes a recording/playing button, an accepting button, a rejecting button, an attribute-correcting button and a value-correcting button.
Preferably, the search unit is a search software.
In accordance with further another aspect of the present invention, a speech recognition method is provided. The speech recognition method includes steps of (a) receiving a speech from a user and recognizing the speech for generating a plurality of recognition results, (b) displaying the recognition results for the user to confirm/correct the recognition results, and (c) searching for a desired datum based on the confirmed/corrected recognition results.
Preferably, the recognition results are shown simultaneously.
Preferably, the recognition results are shown one by one.
Preferably, the step (b) is performed by receiving a next speech from the user.
Preferably, the step (b) is performed by means of a control device.
The above objects and advantages of the present invention will become more readily apparent to those ordinarily skilled in the art after reviewing the following detailed descriptions and accompanying drawings, in which:
The present invention will now be described more specifically with reference to the following embodiments. It is to be noted that the following descriptions of preferred embodiments of this invention are presented herein for the purposes of illustration and description only; it is not intended to be exhaustive or to be limited to the precise form disclosed.
Please refer to
Preferably, the locking device 24 is a button, a touch screen or a remote controller. When the locking device 24 is a touch screen, the touch screen may also serve as the displaying device 23. Moreover, the storage device 25 is preferably a register. The database 26 is preferably a memory, a flash disk, a hard disk or a remote server. Any kinds of data can be searched via the speech recognition system 2 described above, such as the flight timetable, the stock information, etc.
Please refer to
Referring now to
The speech recognition method and system described above have the following advantages.
1. The recognition results are shown on the displaying device 23 in the format of “attribute-value”. Therefore, it is easy for the user to identify which fields are still empty. That is, the user knows which speech he should input next without the questioning from the system.
2. The way of locking known values is adopted to eliminate the occurrence of incorrect speech recognition. After the user inputs his speech, the recognition results will be shown in corresponding fields. The correct values can be selected either by keeping the correct values or by deleting the incorrect values. After that, the correct values kept are locked and regarded as known values that are unchangeable. The next speech from the user can only change the fields that are not locked. Thus, the recognition range can be narrowed down. This not only enhances the rate of recognition but reduces the time required for the speech recognition.
3. The user can input more than one attribute at a time by the way of natural language.
4. The recognition range can be narrowed down when a part of the values for the fields is known.
5. The speech from the user can be re-recognized when a part of the values for the fields is known.
6. The desired datum can be searched automatically by the system without completely filling the fields.
Please refer to
The input device 53 is used for receiving a speech from a user. The speech recognition device 54 performs speech recognition based on a lexicon. The language understanding device 55 performs language understanding based on a grammar rule to generate a plurality of recognition results. The lexicon and the grammar are generated from processing the digital data or the video programs of the storage/receiving device 51 (step 52). The interactive meaning confirmation/correction software 56 is used for confirming/correcting the recognition results. The displaying device 58 is used for displaying the recognition results one by one on a specific region thereof. The keypad 59 is used for the user to confirm/correct the recognition results. Alternatively, the keypad 59 can be replaced with a remote controller or a personal digital assistant. The search software 57 is used for searching the storage/receiving device 51 based on the confirmed/corrected recognition results so as to find out the corresponding digital data or video programs.
The titles of the digital data or video programs being stored or received in the storage/receiving device should be classified in advance according to their attributes. For instance, “You are not alone” by “Michael Jackson” is classified as the value for the attribute of “song”, and the value for the attribute of “singer” is “Michael Jackson”. The program “CNN Live Today” is a value for the attribute of “program name”, the corresponding value for the attribute of “program category” is “news program”, the corresponding value for the attribute of “radio station” is “CNN”, and the corresponding value for the attribute of “time” is “AM 10-12”.
During the search, the user only needs to use daily sentences. For example, the user speaks “turn to CNN Live Today” or “You are not alone by Michael Jackson” In this way, the unnaturally hierarchical instructions, such as speaking “television”, “news program”, and finally the program name “CNN Live Today” in turn, are unnecessary anymore.
The corresponding lexicon and grammar generated from processing the classified titles of the digital data or video programs will serve as the basis of the speech recognition and the language understanding. Furthermore, the speech recognition device 54 and the language understanding device 55 can be combined into a single component.
After the speech from the user is received by the interactive speech recognition understanding device 55, it is interpreted as the “attribute-value” format in pairs by the speech recognition device 54 and the language understanding device 55, even if the user doesn't speak the attribute. For instance, when the user speaks “You are not alone by Michael Jackson” without speaking “singer”, an “attribute-value” pair “singer-Michael Jackson” will be shown on the displaying device. Many “attribute-value” pairs can be generated from a single sentence spoken by the user. Finally, the erroneous meaning is corrected or the correct meaning is confirmed through the interactive meaning confirmation/correction software 56. The speech recognition method for this preferred embodiment will be illustrated in detail as follows.
1. The speech recognition method for this preferred embodiment is designed for confirming/correcting an “attribute-value” pair at a time. In this way, an “attribute-value” pair is shown on a specific region of the displaying device 58, so that the user could still watch the programs normally. In addition, the interactive confirmation and correction can be made easily by using the keypad 59 which consists of five buttons.
2. Only one “attribute-value” pair is shown on the displaying device 58 at a time. Moreover, the keypad 59 consisting of five buttons is provided for interacting with the speech from the user.
3. Please refer to
The recording/playing button: The speech section from the user corresponding to the shown “attribute-value” pair could be played when the recording/playing button is pressed softly. The re-recording function could be performed when the recording/playing button is pressed heavily or lastingly so as to re-confirm/re-correct the “attribute-value” pairs.
The accepting button: The shown “attribute-value” pair are accepted when the accepting button is pressed softly, and then a next action is proceeded. The next action is to show the next “attribute-value” pair that are not confirmed/corrected yet for interaction with the user, if any.
The rejecting button: The shown “attribute-value” pair are rejected when the rejecting button is pressed softly, and then a next action is proceeded. The next action is to show the next “attribute-value” pair that are not confirmed/corrected yet for interaction with the user, if any.
The attribute-correcting button: A new attribute in another Top-N candidate “attribute-value” pair is corrected and selected when the attribute-correcting button is pressed softly. The re-recording function could be performed and then a new attribute in another possible “attribute-value” pair is identified when the attribute-correcting button is pressed heavily or lastingly.
The value-correcting button: A new value in another Top-N candidate “attribute-value” pair is corrected and selected when the attribute-correcting button is pressed softly. The re-recording function could be performed and then a new value in another possible “attribute-value” pair is identified when the value-correcting button is pressed heavily or lastingly.
If there are a plurality of “attribute-value” pairs, the displaying sequence therefor is determined by the system based on an intelligent judgment thereof instead of the sequence of the speech. The consideration for determining the displaying sequence for the “attribute-value” pairs is based on an operation convenience for the user. For instance, the interaction should be highly natural and times for pressing the buttons should be less.
The search could be performed after any of the “attribute-value” pairs is confirmed/corrected. Meanwhile, whether the confirming/correcting process for the unconfirmed/uncorrected “attribute-value” pairs needs to proceed or not is determined automatically by the system. In addition, the search results (the amount or the respective items) could be shown on the displaying device 58 for being consulted.
Referring now to
The function of human-machine interface is provided in the interactive speech recognition understanding device of this preferred embodiment, which is able to search mass information rapidly and effectively. This preferred embodiment could be applied to devices with a small-scale screen, for example, a small digital data storage/playing device such as the MP3 player, the smart phone and so on. Also, this preferred embodiment could be applied to the device with a large-scale screen. The characteristic of this preferred embodiment lies in that only a small part of the screen is used as the communication interface for speech understanding, so that the user could still watch the program normally. For example, it could be applied to the control for the television, the program selection, the adjustment for the video quality, etc. Furthermore, it could also be applied to the control for the video recorder, such as setting the recording time, playing the pre-recorded program and so on, as shown in
Accordingly, the present invention can effectively solve the problems and drawbacks in the prior art, and thus it fits the demand of the industry and is industrially valuable.
While the invention has been described in terms of what is presently considered to be the most practical and preferred embodiments, it is to be understood that the invention needs not be limited to the disclosed embodiments. On the contrary, it is intended to cover various modifications and similar arrangements included within the spirit and scope of the appended claims which are to be accorded with the broadest interpretation so as to encompass all such modifications and similar structures.
| Number | Date | Country | Kind |
|---|---|---|---|
| 094102062 | Jan 2005 | TW | national |
