DE102009033384A1 - Method for operating a device and device - Google Patents

Abstract

The method involves converting language information into textual information by a language recognizer (50), and controlling the device by the converted textual information. The subset is selected from a data base (20) assigned to the device for run-time of the device. The wordbook for the language recognizer is provided from the subset. The subset is formed from the data elements. An independent claim is also included for a device comprises a language recognizer for converting language information into textual information.

Description

Field of the invention

The The invention relates to a method for operating a device according to the preamble of claim 1.

The The invention further relates to a device according to the preamble of claim 9.

Background of the invention

A Device in the sense of this invention is an electronic system, which is suitable, at least partially served via a voice input to become. Which includes in particular a navigation system or a music system for motor vehicles.

Out the book "Multislot Inputs in voice-operated navigation devices "by Steffen Rhinow, VDM Verlag Dr. Miller, Saarbrücken, published in 2008, a navigation device is known in which an address out of town, street and house number over the language is transferred to the navigation device. The navigation device converts the spoken address by means of a speech recognizer in a text information around, from which a destination for a route guidance is determined. If the text information is not unique, it is done the selection of the destination by a dialogue with the user.

In The book by Steffen Rhinow introduces the principle of speech recognition explained in the essential steps as follows: An analog voice signal will be first considering digitized from the Nyquist-Shannon sampling theorem. thereupon there is a preprocessing of the digitized speech signal. The preprocessing includes spectral analyzes for small periods of time, in where the speech signal does not change significantly, for example, for periods of time of ten milliseconds duration. From the information thus obtained Feature vectors extracted. In addition, often find a noise reduction and detection of the beginning, pauses and end of the speech signal. thereupon is based on a language model, a dictionary and an acoustic Model according to a most likely recognized feature vector sequence searched. As a result, usually one is after detection probabilities ordered list (so-called NBest list or hit list) possible Information supplied. The dictionary contains all to recognize words together with the corresponding phoneme notation. The dictionary can by using so-called G2P programs (G2P, Graphem To Phonem) to be created. These programs build on the grapheme of words a dictionary with the corresponding phonemes.

Summary of the invention

task The present invention is a method for operating a To develop a device and a device of the type mentioned in the introduction, in particular speech recognition in a simple and cost-effective manner to update and to improve the detection probabilities.

The The object is achieved by a Method with the features of claim 1 and by a device solved with the features of claim 9. Particular embodiments are in the subclaims, the description and the figure disclosed.

Under a device in the context of this invention is both a single Device as also a multifunction device understood that operated in a context of multiple contexts becomes.

A Voice information is an analog voice signal containing information contains. A textual information is a (machine-readable) string that the Contains information.

at the method according to the invention becomes the dictionary for the Speech Recognizer created at runtime of the device, that is, at a time to which the device is turned on or a voice input mode the device is activated. The dictionary will be one of the Device associated database created by a subset, the Also, the entire database may contain extracted from the database and in a dictionary the speech recognizer is converted. The conversion is done by a G2P program.

The method no longer requires a database independent of the database because it is generated from the database. The dictionary is thus always up to date. In conventional voice operated devices, the dictionary is independent of the database. Since the database is regularly updated more often than the dictionary of the speech recognizer, inconsistencies can occur if the database receives new words that are not yet part of the dictionary. If the user wants to avoid such inconsistencies, he would have to keep both the dictionary and the database on the same level. This is not always possible due to the lack of availability of new dictionaries and would be associated with additional costs. By contrast, in the method according to the invention and the device according to the invention, it is only necessary to update the database. The dictionary then automatically has the same status as the database. Thus, the invention To keep the method and the device according to the invention inexpensive current. In addition, inconsistencies due to an outdated dictionary and a more up-to-date database than the dictionary are avoided.

In an embodiment The database stores files that contain data elements. The subset is made up of data elements. The subset contains it only those data elements needed to operate the device. This will reduce processing power and memory requirements during conversion the subset in the dictionary, saving the dictionary and speech recognition is reduced.

In a further embodiment the selection takes place the subset from the database using a filter. The filter is useful as a computer algorithm executed which can be easily adapted. You can get the computer algorithm vividly imagine as follows: "Search all the data elements out of the database that meet a certain requirement, and add this Add data elements to the subset. "In a further embodiment the method is performed recursively in a multi-part information, wherein the filter for every loop is adjusted. A multi-part information exists from several pieces of information that are required to a particular one To fulfill the task For example, the sub-information "city", "street" and "house number" of the information "address" at a Destination entry in a navigation system.

In a further embodiment the device is a navigation system and the database is one Navigation database containing a digital road map. Conveniently contains the text information at least one destination for a route guidance.

In a further embodiment the device is eleven music system and the database is a music database. Conveniently contains the text information at least one song.

The inventive device has the same advantages as the inventive method, so that on the above statements is referenced.

Short description of the drawing

A preferred embodiment of the invention is based on the single 1 explained in more detail.

Detailed description the drawing

1 schematically shows a structure of a device according to the invention, as a navigation system 10 is executed in a motor vehicle.

The navigation system has a database designed as a navigation database 20 , an application system 30 , a dialog control 40 , a speech recognizer 50 and a voice output 60 on.

Database 20 includes a digital road map. Database 20 Contains files that store data items that uniquely characterize a stretch of road. In particular, the data elements include the name of the city in which the road section is located, the name of the road in which the road section lies, house numbers located in the road section.

The application system 30 controls a route guidance to a destination, in particular a route calculation between a current location of the motor vehicle and the destination performs and determines driving instructions.

The dialog control 40 Controls a dialog between a user and the application system 30 ,

The navigation system 10 has a speech recognizer 50 on, which provides the user's voice information in one for the application system 30 converts processable text information.

The navigation system 10 has a voice output 60 on which the user driving recommendations, prompts, etc. can be given.

A block arrow 70 indicates an input by the user and a block arrow 80 an output by the speech output. arrows 90 indicate a data exchange between the components of the navigation system 10 , The data flow takes place in the direction of the arrow.

The following is the dialogue control 40 controlled interaction of the speech recognizer 50 with the database 20 explained in more detail.

After the user has activated the navigation system and voice input, the dialog control creates 40 using a filter in a default setting, a subset from the database 20 , The basic setting of the filter depends on the database 20 from. Includes the database 20 a digital map of a country, the filter could be defined as: "Extract all cities and pass the cities into the subset." Includes the database 20 digital maps of several countries, such as all Western European countries ", the filter could be defined as follows: "Extract all countries and transfer the countries into the subset." There are also other basic settings for the filter conceivable, which can be optionally set by the user individually. In the present embodiment, the database includes 20 a digital road map of the Federal Republic of Germany. In addition, the user has set the navigation system so that the filter by default all city names from the database 20 extracted.

The subset with all city names is stored in a suitable memory of the speech recognizer 50 saved. The dialog control then processes 40 using a G2P program stored in a suitable memory of dialog control 40 is stored, the subset. This processing turns the subset into a dictionary for the speech recognizer 50 created, which contains the city names with their genomes and phonemes. The dictionary is stored in a suitable memory of the speech recognizer 50 stored and is available for voice input by the user.

The user then enters an address or part of an address via his language, depending on whether the speech recognizer can process multi-part information (such as the entire address with city, street and house number) or in one pass only one-piece information (either city or street or house number). In the present case, the speech recognizer is in interaction with the dialogue control 40 and the database 20 able to process multi-part information. The speech recognizer 50 is also designed as a serial speech recognizer, that is, he divides the multi-part information into several time slots, so-called slots. Each slot is assigned a piece of information.

For example the first slot becomes the city, the second slot the street and the third slot assigned the house number. The slots are timed evaluated one after the other.

With the dictionary for the speech recognizer generated from the basic setting of the filter 50 Now the part information of the first slot will be recognized. This subinformation is the city name "Berlin". Since this city name is in the dictionary, the speech recognizer recognizes 50 this city with a high detection probability. The speech recognizer 50 is designed so that when it exceeds a predetermined high detection probability and / or a predetermined difference to the hit with the second highest detection probability of a clear result and continues without a confirmation or selection by the user.

The Subset is deleted.

Now begins a second loop of recursive address recognition. For this will be first the filter is redefined and adapted to the second loop. The Definition is now as follows: "Search from the database all Street name of the City Berlin and transfer the street names into the subset. "Subsequently, it becomes as with the city names using the G2P program, a street name dictionary from Berlin for the Speech recognizer generated. This dictionary stands for the evaluation of the second slot, which contains about the street name "Kurfürstendamm" as part information to Available.

In a third, not closer executed, loop the third slot is evaluated with the house number.

After this all three slots have been evaluated, in the best case the Address of exactly one destination, the (or their for the navigation system suitable geographic coordinates) used as the destination for the route guidance becomes. If multiple addresses are considered destinations, is the user is prompted to select an address. This selection is made For example, by manual input via a touch screen or voice-controlled.

The above-mentioned memory of the database 20 , Dialogue control 40 and the speech recognizer 50 can be realized in several or in a storage medium. If only one storage medium is used, the different memories in different partitions can be realized.

For the different addressed controls can several control processors or even a control processor used become. If only one control processor is used, its performance becomes through a suitable tax management on the different ones Processes split.

LIST OF REFERENCE NUMBERS

10

navigation system

20

Database

30

application system

40

dialog control

50

spokesman Kenner

60

speech

70

block arrow

80

block arrow

90

pipe

Claims (16)

Method for operating a device, in which voice information is transmitted by means of a speech recognizer ( 50 ) is converted into text information and the text information for controlling the device is processed, characterized in that at runtime of the device from a database associated with the device ( 20 ) selected a subset and from this subset a dictionary for the speech recognizer ( 50 ) is created. Method according to claim 1, characterized in that in the database ( 20 ) Files are stored, that each file contains data elements and that the subset is formed of data elements. Method according to claim 1 or 2, characterized in that the selection of the subset from the database ( 20 ) by means of a filter. Method according to one of the preceding claims, characterized characterized in that the method in a multi-part information recursively performed with the filter adjusted at each level. Method according to one of the preceding claims, characterized in that the device is a navigation system ( 10 ) and the database ( 20 ) is a navigation database. Method according to claim 5, characterized in that in that the text information contains at least one destination for a route guidance. Method according to one of claims 1 to 4, characterized in that from the device a music system and the database ( 20 ) is a music database. Method according to claim 7, characterized in that that the text information contains at least one music title. Device with a speech recognizer ( 50 ) for converting a speech information into a text information and a controller for controlling the device in dependence on the text information, characterized in that the device is designed, at runtime of the device from a database associated with the device ( 20 ) select a subset and from this subset a dictionary for the speech recognizer ( 50 ) to create. Device according to claim 9, characterized in that in the database ( 20 ) Files are stored, that each file contains data elements and that the subset is formed of data elements. Apparatus according to claim 9 or 10, characterized by a filter for selecting the subset from the database ( 20 ). Device according to one of claims 9 to 11, characterized that the device is designed, multi-part information recursively process and filter at each stage of the recursion adapt. Device according to one of claims 9 to 12, characterized in that the device as a navigation system and the database ( 20 ) is designed as a navigation database. Device according to claim 13, characterized in that in that the text information contains at least one destination for a route guidance. Device according to one of claims 9 to 12, characterized in that the device as a music system and the database ( 20 ) is designed as a music database. Device according to claim 15, characterized in that that the text information contains at least one music title.

Images