WO2009109169A1

WO2009109169A1 - Method for operating a navigation system

Info

Publication number: WO2009109169A1
Application number: PCT/DE2009/000273
Authority: WO
Inventors: Jan-Georg Fritsche
Original assignee: Navigon Ag
Priority date: 2008-03-03
Filing date: 2009-02-27
Publication date: 2009-09-11
Also published as: DE102008027958A1; EP2250464A1

Abstract

The invention relates to a method for operating a navigation system comprising a receiving device in which an acoustic address input, that consists of several input components, can be registered. Said input components of the address are analysed by a voice recognition module and depending on the result of the voice recognition analysis, at least one geographic location, that is defined by an address having several address components, is selected from a data base for further processing. At least one input component is temporarily stored as an acoustic data set in a main memory and the temporarily stored acoustic data set is downloaded later from the main memory.

Description

Method for operating a navigation system

The invention relates to a method for operating a navigation system with regard to the acoustic speech recognition according to the preamble of claim 1.

Known navigation systems, which may be, for example, mobile or fixed navigation devices, serve the user to compute a route from a starting point to a destination point to subsequently issue maneuvering instructions for following that route. The basis of route planning is in many cases the input of an address by the user. To enable the entry of addresses, the navigation system may be equipped with an alphanumeric keyboard or a touch screen, at which the user inputs the corresponding address components, such as the city name, the street name and the house number. In addition, navigation systems with speech recognition analysis are becoming more widespread. In these systems with speech recognition analysis, a voice recognition module is present in which acoustically entered operator inputs, in particular acoustically entered Address inputs, analyzed and converted into electronically processable data.

Depending on the quality of the speech recognition module, very good hit rates for speech recognition are already achieved. However, since speech recognition, in particular also with regard to the very different ways of speaking of the various users, can not be assumed to be a hundred percent hit rate, the known navigation systems are additionally equipped with correction procedures. If, for example, the user has entered the name of the town, the street name and the house number, then the street is recognized by a street vocabulary that has been generated from all the streets of all recognized cities. However, this raises the problem that in cases where the city name was not recognized correctly, the correct recognition of the street name is often impossible, since the corresponding street in another city is highly likely to be nonexistent. If the user now corrects the city name in the correction workflow, the street name must then be entered again acoustically or via the keyboard, even though the street name has already been entered acoustically and has not changed in the meantime. This leads to lengthy and complicated procedures in the correction of address input.

Based on this prior art, it is therefore an object of the present invention to simplify and improve the possibilities for correcting an address input.

This object is achieved by a method according to the teaching of claim 1.

Advantageous embodiments of the invention are the subject of the dependent claims. The inventive method is based on the basic idea that at least one input component of the acoustic address input, for example, the acoustically entered street name, as an acoustic data set in a main memory of the navigation system is cached. Caching allows this acoustic data set to be reloaded from memory and re-analyzed at any later time. This reloading of address inputs already spoken by the user is advantageous, in particular with regard to the correction of address information. If the system has not correctly recognized the city name, for example, after the user has corrected the city name, for example by inputting a keyboard, the voice analysis of the further address components can be stored by reloading the corresponding acoustic data records in which the associated input components of these address components are stored are again performed in the voice recognition module without the user having to re-address these address components.

According to a preferred variant of the method, it is provided that the user accepts at least two input components, namely a first input component and a second input component, for example a city name and a street name, into the receiving device of the navigation system and this acoustic recording is analyzed in the voice recognition module. In addition, at least one input component, for example the street name, is buffered as an acoustic data record in the main memory. Now, if the correction of the first input component is necessary, for example by incorrectly recognized city names are corrected in the speech recognition analysis, then the cached acoustic data set containing, for example, the street name can be loaded from the main memory and analyzed again in the speech recognition module. Of particular advantage is the method according to the invention, when all input components of the address input, in particular the city name and / or the street name and / or the house number, are recorded acoustically together and behind each other, regardless of an in each case interspersed voice recognition analysis. In the case of these so-called "one-shot input solutions", the user can then acoustically respond to all the necessary inputs in a very short time, for example by issuing a corresponding input request for inputting the respectively assigned address component between the individual inputs Address input are recorded and cached as acoustic data sets in the main memory, further acoustic speech input by the user to specify the address is no longer necessary because the corresponding information can be made available at any time by loading the cached acoustic data sets again.

The speech recognition analysis may optionally be performed after the complete input or in the background, that is, as a concurrent thread, in parallel with the speech input.

As far as all input components of the address input are recorded acoustically, the various input components should each be cached as separate acoustic data records, so that the different parts of the address input can be accessed effectively and purposefully. If, for example, the name of the city, the street name and the house number are spoken by the user, a separate acoustic data record should be generated both for the city name and for the street name and for the house number and buffered in the main memory.

The form in which the acoustic data of the acoustically recorded address entries are temporarily stored is basically arbitrary. Particularly suitable are phoneme vectors that have a storage with allow relatively small memory requirements and also realize an effective speech recognition analysis.

If the method according to the invention is implemented with regard to the correction of a city name, then it is particularly advantageous if, after correction of the city name, a vocabulary list of all street names stored for this city is created. Subsequently, the cached in the working memory acoustic data in which the street name is cached, loaded into the speech recognition module and compared with regard to the newly created vocabulary list with all existing in the city of the corrected city name street names. Re-speaking the street name after the correction of the city name is therefore no longer necessary.

The method according to the invention can also be used with regard to the correction of the street name. If the street name has been corrected, then a vocabulary list can be created with all house numbers present on this street, and in the subsequent speech recognition analysis the stored acoustic dataset can be compared with these house numbers of the corrected vocabulary list.

Various aspects of the invention are illustrated schematically in the drawings and are explained below by way of example.

Show it:

Fig. 1 shows the input mask for entering an address;

FIG. 2 shows the input mask according to FIG. 1 during voice input of the city name, the street name and the house number; FIG.

3 shows the mask for indicating the address input after carrying out the first speech recognition analysis;

4 shows the list for displaying further hits of the first speech recognition analysis; FIG. 5 shows the input mask for correcting the address input after the speech recognition analysis; FIG.

FIG. 6 shows the input mask according to FIG. 5 during the re-entry of the city name; FIG.

FIG. 7 shows the hit list for displaying the city names after the second speech recognition analysis; FIG.

8 shows the input mask for the correction of the street name with already corrected street name;

9 shows the input mask for correcting the address by alphanumeric keyboard input.

In the following, the method according to the invention will be explained with reference to an example in which the user would like to enter the address "Berliner Platz, 1 1, Würzburg" as the destination address in the navigation system The navigation system then loads the vocabulary of all place names in the selected country, namely Germany, whereupon the navigation system asks the user to enclose the name of the city, whereupon the user speaks the word "Würzburg" into the recording device of the navigation system, for example a microphone. If, for example, there were disturbing background noises during this input of the city name, the correct city name can not be recognized due to the poor recording quality, so that the correct city name "Würzburg" is not included in the best hits of the speech recognition analysis the best n hits of the speech recognition analysis for the city name a common vocabulary for the subsequent speech recognition analysis with regard to the street name is created, the street name is "Berliner Platz" in This street name vocabulary to be used for the further speech recognition analysis does not exist.

If the user is now prompted to enclose the street name and, as shown in FIG. 2, enters the street name "Berliner Platz", the correct street name can not be included due to the street name vocabulary that does not contain the street name "Berliner Platz". not be recognized from the outset. After the voice input of the acoustically spoken street name "Berliner Platz", a phoneme vector for this voice input of the corresponding input component was generated and temporarily stored in the main memory of the navigation system.

The user is then prompted by the navigation system to also enter the house number, whereupon the user acoustically enters house number "1 1." After performing the speech recognition analysis for all three input components that were entered acoustically by the user, the speech recognition analysis is the best match the address shown in Fig. 3. However, due to the poor quality of the image when entering the city name, an incorrect address, namely "Wolzburg, Borbitzweg, 1 1" was detected. The result obtained on the basis of the first speech recognition analysis thus contains neither the correct location nor the correct street. Only the house number was recognized correctly.

Subsequently, the user is asked whether the displayed result of the address input is correct. Since the result is not correct, the list shown in FIG. 4 is subsequently displayed, from which the further hits of the first speech recognition analysis can be seen and displayed to the user. If the desired address is also not included here, the user can start a correction workflow whose input mask is shown in FIG. In this correction input mask, the user can first correct the city name, for example by renewed acoustic input of the city name. As 6, the city name recognized after the first speech recognition analysis is first deleted from the address input, and the user is shown the hit list after the second speech recognition analysis for the city name, as shown in FIG. Due to the improved acoustic conditions, the correct city name "Würzburg" has now been recognized and this correct city name can then be accepted and read into the input mask.

Then, as shown in Fig. 8, the correction of the street name is started. When starting the correction of the street name, the navigation system first loads the phoneme vector of the first speech input, which is temporarily stored in the working memory for the street name, in which the street name is stored. This cached phoneme vector is now matched against the street name vocabulary generated due to the changed city name. This city name vocabulary for the city of Würzburg now also contains the "Berliner Platz" as a possible street and is therefore recognized correctly, so that the street name "Berliner Platz" is automatically entered by the navigation system into the correction input mask according to FIG displayed for selection. Since now all address components are entered correctly, the user can take over this address in his route planning by the navigation system.

FIG. 9 shows an alphanumeric input mask with which the user can overwrite all incorrectly recognized address components by manual keystrokes. Since the address has been recognized correctly by the speech recognition method described, a route planning to the desired destination can be made by pressing the button labeled "Start navigation".

Claims

claims

1 . Method for operating a navigation system with a recording device, at which an acoustic address input, which consists of several input components, can be recorded, wherein the input components of the address are analyzed with a voice recognition module, and depending on the result of the speech recognition analysis at least a geographical location, which is defined by an address with multiple address components, is selected from a database for further processing, characterized in that at least one input component is buffered as an acoustic data record in a working memory, wherein the buffered acoustic data record is loaded later from the working memory.

2. The method according to claim 1, characterized in that at least a first and a second input component are analyzed after the acoustic recording in the speech recognition module, wherein at least the second input component is buffered as an acoustic data set in a main memory, and wherein the buffered acoustic data set after a correction of the first Input component is loaded from memory and re-analyzed in the speech recognition module.

3. The method according to claim 1 or 2, characterized in that all input components of the address input, in particular the city name and / or the street name and / or the house number, are recorded acoustically in succession.

4. The method according to any one of claims 1 to 3, characterized in that the acoustically recorded input components of the address input are stored as a separate separate acoustic data sets.

5. The method according to any one of claims 1 to 4, characterized in that the acoustic data sets are cached in the manner of phoneme vectors.

6. The method according to any one of claims 1 to 5, characterized in that after correction of the city name a vocabulary list of all street names stored for this city is created, wherein in the subsequent speech recognition analysis for recognizing the user-spoken street name of the street name associated, cached Acoustic data set is compared with the street names of the vocabulary list.

7. The method according to any one of claims 1 to 6, characterized in that after correction of the street name a vocabulary list of all stored on this street house numbers is created, wherein in the subsequent speech recognition analysis for detecting the user-spoken house number of the house number associated, cached acoustic data set is compared to the house numbers of the vocabulary list.