GB2402855A

GB2402855A - Multiple language text to speech processing

Info

Publication number: GB2402855A
Application number: GB0313710A
Authority: GB
Inventors: Premalasen Rajalingham
Original assignee: Seiko Epson Corp
Current assignee: Seiko Epson Corp
Priority date: 2003-06-12
Filing date: 2003-06-12
Publication date: 2004-12-15
Also published as: GB0313710D0

Abstract

At present, a user of a mobile telephone initialises the same by selecting the operating language. This includes selecting the language of the TTS engine. This selection needs to be based upon the expected input text data. In more and more cases the text data is in a different language to that expected by the TTS engine. Hence, the TTS engine either outputs nonsense or is not able to function. Thus, the present invention relates to a method of processing text data in a TTS engine comprising: receiving said text data; analysing said text data to determine a predominant language in which said text data is written; selecting said predominant language for said TTS engine; and outputting speech from said TTS engine in said predominant language.

Description

A method of processing text data in a Text-to-Speech engine, a semi-

conductor chip and a mobile telephone, each arranged to effect said method The present invention relates to a method of processing text data in a Text-to-Speech (TTS) engine, a semi-conductor chip and a mobile telephone, each arranged to effect said method.

A TTS engine converts text to speech. TTS engines are employed in many applications including mobile telephones, computers, transport announcements and other such applications. In many applications, more than one language is supported. In particular, manufacturers of mobile telephones will include many TTS engines, each one relating to one of the common languages of the expected market. Thus, mobile telephones in Europe may include 15 or more TTS engines, each corresponding to one of the European languages. As globalisation progresses further with the reduction of technological trade barriers, the same mobile telephones will hitherto be marketed in not only Europe but also North and South America, Asia and the Baltic states. Thus, the number of languages to be supported will increase more and more. In addition, users, particularly in Europe, are becoming not only bilingual but also multilingual, switching and borrowing between languages in the same sentence.

At present, a user of a mobile telephone initialises the same by selecting the operating language. This includes selecting the language of the TTS engine. This selection needs to be based upon the expected input text data. In more and more cases, the text data is in a different language to that expected by the TTS engine. Hence, the TTS engine either outputs nonsense or is not able to function.

Thus, the present invention seeks to overcome these problems by providing a method of processing text data in a TTS engine which is independent of the pre-selection of language.

According to the present invention, there is provided a method of processing text data in a TTS engine comprising: receiving said text data; analysing said text data to determine a predominant language in which said text data is written; selecting said predominant language for said TTS engine; and outputting speech from said TTS engine in said predominant language.

The present invention also relates to a semi-conductor chip incorporating a TTS engine, having means coupled to said TTS engine for analysing text data to determine a predominant language in which said text data is written, said semi-conductor arranged to effect the method as defined in the attached claims.

Finally, the present invention relates to a mobile telephone having such a semi- conductor chip.

Embodiments of the present invention will now be described by way of further example only and with reference to the accompanying drawings, in which: Figure 1 is a schematic diagram of a method of processing text using a TTS engine

according to the prior art;

Figure 2 is a schematic diagram of a method of processing text using a TTS engine according to a preferred embodiment of the present invention; Figure 3 is a schematic diagram of a neural network used to recognise the language of the text; Figure 4 is a schematic diagram of an alternative method of processing text; Figure 5 is another schematic diagram of another alternative method of processing text; Figure 6 is a schematic diagram of a semiconductor chip according to the present invention; and Figure 7 are schematic diagrams of various applications in which a TTS engine may be embedded.

Figure 1 is a schematic diagram of a method of processing text using a TTS engine according to the prior art. A receiver 2 is arranged to receive text input and this is passed to a message decoder 4. The message decoder extracts any control code in the text input and so isolate the text data. A user has already initialised the processing method by pre- selecting the language to be used. Message decoder implements the selection by enabling the respective TTS engine through the use of switch 6. In this example of the prior art, language A has been selected. The message decoder 4 passes the text data to the pre-selected and enabled TTS engine 8a. The TTS engine converts the text to speech and outputs the same by output means 10.

In the prior art, a plurality of TTS engines 8a, 8b. . . 8x are provided; each TTS engine relating to a particular language. Thus, a large number of TTS engines need to be provided which adds to the complexity and cost. Moreover, if the text input is not in the pre-selected language, then the selected TTS engine may not be able to process and output suitable speech.

Figure 2 illustrates a method of processing text using a TTS engine according to a preferred embodiment of the present invention. The preferred embodiment includes receiver 2 and message decoder 4 as well as TTS engines 8a, 8b... 8x and output means 10.

However, the present invention also includes language recognition means 12. Language recognition means 12 receives the text data from message decoder 4. Language recognition means 12 analyses the text data to determine the predominant language of the input text.

The result of this recognition is passed to selection means 14. Selection means 14 selects the corresponding TTS engine according to the recognized predominant language. Selection means 14 thus enables the particular TTS engine. The TTS engine processes the text data to determine the corresponding speech and output means 10 outputs the speech.

Language recognition means 12 can effect it's functions in a number of ways. In the first instance, the language recognition means could be arranged to receive the control code which has been separated by the message decoder 4. That control code could also include details of the predominant language in which the text data is written. The language recognition means having received the predominant language information from the control code, can instruct the selection means 14 to select and enable the respective TTS engine.

Alternatively, language recognition means 12 could be arranged to classify patterns within the text data. Using neural network logic the most likely language can be selected on the basis of the classification analysis.

A neural network can be used to recognise the language of the text data. The neural network can be trained to analyse the text data and map the text data to a specific language.

Figure 3 is a schematic diagram of an example of a feedforward neural network. A feedforward neural network is a network where the input text is passed from an input layer 20 of neurons through a hidden layer 22 to an output layer 24 of neurons. The function of the hidden layer is to process the information from the input layer. The hidden layer comprises a set of rules which is used to determine the output in relation to the input. The neural network can be trained using supervised learning whereby it is able to analyse input text and classify the input text to target languages based on a set of pre-learnt rules.

Finally, language recognition means 12 could be arranged to identify specific characters or most frequently used words corresponding to particular languages. For example, the characters a, a, a, a, a, a, He could comprise the specific characters. Most frequently used words could include the word "and" which in French is "et", in German "und" in Spanish "y" and in Italian "e".

The language recognition means 12 can use each of these three recognition methods.

Moreover, the processing method can be pre-initialised to select the language recognition method used andlor the specific characters or specific words to use or the user can be given the ability to select the language recognition method to be used or the characters or the most frequently used words.

Sometimes, text data includes phrases from another language. For example English text often includes phrases in another language such as "deja-vu" or "gesun&eit" or "buenas noches". The present invention also enables text including these words which are not in the predominant language to be converted to speech in English or to retain their meaning, in the other secondary language which is used.

As shown in Figure 4, language recognition means 12 can be arranged to identify minor portions of the text data which relate to words not in the predominant language used for the rest of the text data. The language recognition means breaks up the sentence into the predominant portion and the minor portion in the secondary language. The language recognition means can recognise the language used for the minor portion and instruct the selection means 14 to enable a different ITS engine corresponding to the secondary language used for the minor part of the text data. The selection means 14 can then pass the text data in the predominant language to the enabled predominant TTS engine and the minor portion relating to the secondary language to the secondary TTS engine. Output means 10 can then re-integrate the minor portion in the secondary language back into the text data which is in the predominant language and output the same seamlessly.

Alternatively as shown in Figure 5, the language recognition means 12 could break the sentence up into the predominant portion in the predominant language and the minor portion - in the secondary language. The minor portion is passed to an automatic translation module for translating the minor portion relating to the secondary language into the predominant language. The sentence is then reconstructed prior to passing the text data to the TTS engine relating to the predominant language. Thus, the speech being output is all in the i predominant language with that minor part also being presented in the predominant language. Thus, phrases such as "buenas noches" when used in a English sentence would be translated to an output as "good night".

The method for processing text data according to the present invention is independent of the pre-selection of the language. Thus, the processing method is particularly advantageous in applications which are likely to receive text data in different languages.

This is particularly important in transport announcements as integrated transport across different countries using different languages is becoming more and more common.

The method may most readily be effected using a semiconductor chip whereby the various features and functions are provided either in hardware or software or a combination.

Figure 6 is a schematic diagram of a semiconductor chip 38 according to the present invention. The semiconductor chip comprises language recognition means 12, output means and a microprocessor 40 which comprises the selection means 14 and TTS engines 8a, 8b....8x. Text data may be input to a keypad 42 which is recognised by language recognition means 12. The microprocessor 40 receives the text data and the details of the predominant language from the language recogr. ution means 12. The microprocessor selects the appropriate TTS engine to use. The microprocessor may call upon a memory 44 to obtain the different language rules. The microprocessor having converted the text data outputs the same to output means 10 comprising a digital to analgue converter which is then output as speech.

A semiconductor chip may be defined to provide the alternatives as discussed above and as shown in Figures 4 and 5.

In addition, such a semiconductor chip could be embedded in a number of applications and examples of which are shown in figure 7. Figure 7a comprises a mobile telephone in which such a semiconductor chip effecting the method as discussed above will be particularly important so as to enable multi-language text data to be output correctly. Figure 7b relates to a computer which would obviate the need to select the language to be used by the computer when first setting the computer up. Finally, figure 7c relates to a digital camera which can be used to record and receive text data in many different languages.

The aforegoing description has been given by way of example only and it will be appreciated by a person skilled in the art that modifications can be made without departing from the scope of the present invention.

For example, it is envisage that a single ITS engine could be used with access to multiple databases, each relating to a different language. When the method is implemented on a semiconductor chip, the memory 44 would include the different language rules for each of the languages serviced by the ITS engine.

Moreover, if the microprocessor of the semiconductor chip has sufficient processing resources, then the language recognition means can be omitted and it's functions provided by the microprocessor.

Claims

Claims 1. A method of processing text data in a TTS engine comprising:

receiving said text data; analysing said text data to determine a predominant language in which said text data is written; selecting said predominant language for said TTS engine; and outputting speech from said TTS engine in said predominant language.
2. A method as claimed in claim 1, in which said analysing said text data comprises extracting from said text data, language code data relating to said predominant language.
3. A method as claimed in claim 1, in which said analysing said text data comprises recognising in said text data, key data indicative of said predominant language.
4. A method as claimed in claim 3, in which said key data comprises unique letters specific to said predominant language.
5. A method as claimed in claim 3, in which said key data comprises unique words specific to said predominant language.
6. A method as claimed in claim 5, further comprising initialising said processing of text data by selecting the unique words as the words most likely to be frequently used.
7. A method as claimed in any one of the preceding claims, further comprising identifying a minor portion of said text data which is not written in the predominant language; analysing said minor portion to determine a minor language in which said minor portion is written; selecting said minor language for said TTS engine when converting said minor portion to speech; and said outputting speech from said TTS engine includes outputting all the text data.
8. A method as claimed in claim 7 in which said outputting speech includes outputting all the text data in the predominant language.
9. A semiconductor chip incorporating a ITS engine, having means coupled to said ITS engine for analysing text data to determine a predominant language in which said text data is written, said semiconductor arranged to effect the method as claimed in any one of claims 1 to 8.
10. A mobile telephone having a semi-conductor chip as claimed in claim 9.