Data processing: speech signal processing – linguistics – language – Speech signal processing – Application
Reexamination Certificate
2000-10-17
2003-09-02
Knepper, David D. (Department: 2645)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Application
C704S002000, C704S010000, C704S009000, C704S257000
Reexamination Certificate
active
06615178
ABSTRACT:
TECHNICAL FIELD
The present invention relates to an apparatus and method for translating a speech input in one language into a sentence in other language, and a recording medium having recorded therein a speech translation controlling program.
BACKGROUND ART
Nowadays, as the speech recognition technics have made remarkable advances, apparatuses for translating a recognized speech in one language into a sentence in other language have been developed. For a variety of users, the speech translating apparatus generally uses a recognition engine oriented for every user.
FIG. 1
shows a conventional speech translator
100
. As shown, the speech translator
100
includes an input unit
102
which is a microphone for example, a feature extractor
103
to extract features of a speech from the input unit
102
, a speech recognition collator
104
, an acoustic model memory
105
to store acoustic models which represent phoneme, etc., a word dictionary memory
106
to store a word dictionary which represents correspondences between words and acoustic models, and a grammar memory
107
to store grammar which represents word conjunctions (generally stochastic language models).
The input unit
102
converts a supplied speech into a speech signal, digitizes the speech signal, and then supplies the digitized speech signal to the feature extractor
103
. The feature extractor
103
calculates a string of feature vectors from the digitized speech signal, and supplies the feature vector string to the speech recognition collator
104
.
The speech recognition collator
104
uses the feature vector, the acoustic models stored in the acoustic model memory
105
, the word dictionary stored in the word dictionary memory
106
and the grammar stored in the grammar memory
107
to recognize a word string uttered.
The speech translator
100
further includes a translation block
108
which is supplied with a collation result from the speech recognition collator
104
, a conversion information memory
109
, a sentence corrector
110
, a correction information memory
111
to store correction rule information for use to correct an incorrect sentence, a grammar information memory
112
to store information on example sentence substitution and a result display unit
113
which is a CRT (Cathode-Ray Tube) or a LCD (Liquid Crystal Display) for example.
The translation block
108
roughly translates the collation result by means of the conversion information memory
109
which stores many example sentence information, and supplies the translation result to the sentence corrector
110
. The sentence corrector
110
corrects in detail the translation result from the translation block
108
by the use of the information stored in the correction information memory
111
and grammar information memory
112
.
The result display unit
113
displays the correction result from the sentence corrector
110
as the translation of the supplied speech.
Because of the recognition engine being oriented for unspecified utterers, the speech translator
100
can advantageously be used by any users, while it has difficulty in discriminating for each specified utterer the utterance manner, range of vocabulary and utter's habit which would be constant. Thus, the speech translator
100
has to process even unnecessary speech recognition, since it does not perform speech recognition processing for each utterer.
Also, the speech translator
100
cannot optimize the interaction between the user as an utterer and the translator itself, since it has no user-adaptability or has not any means for saving the utilization frequency and tendency of each user. Thus, the speech translator
100
is likely to incur errors which would not otherwise take place. Furthermore, since the speech translator
100
cannot identify the preference or utilization mode of each user, it does not suit each user at all.
DISCLOSURE OF THE INVENTION
Accordingly the present invention has an object to overcome the above-mentioned drawbacks of the prior art by providing a speech translating apparatus and method, which work at a higher speed and enable optimum interactions between the user and the translator itself, and a recording medium having recorded therein a speech translation controlling program.
The above object can be attained by providing a speech translating apparatus including:
means for extracting features of a supplied speech to provide a feature vector;
means for collating the feature vector from the feature extracting means with a plurality of collation information for speech recognition to recognize the speech and providing a sentence indicative of the recognized speech;
means for translating the sentence indicative of the speech recognized by the speech recognition collating means into a sentence in a language different from that of the supplied speech;
means for correcting the sentence translated by the translation processing means by the use of optimum one of a plurality of correction information;
means for outputting the correction result of the sentence correction means;
means for accumulating history information indicative of the tendency of the sentence recognized by the speech recognition collating means; and
a cumulative learning means for comparing the speech recognition result with the history information accumulated in the history information accumulating means each time the speech recognition collating means outputs the speech recognition result to update the history information accumulated in the history information accumulating means by a cumulative learning function;
the history information accumulating means controlling, based on the accumulated history information, the selection of the collation information for use by the speech recognition collating means and/or selection of the correction information for use by the sentence correction means.
In the speech translating apparatus, history information indicative of the tendency of the sentence recognized by the speech recognition collating means is accumulated, and the speech recognition result is compared with the history information accumulated in the history information accumulating means each time the speech recognition collating means outputs the speech recognition result to update the history information accumulated in the history information accumulating means by a cumulative learning function. The history information accumulating means controls, based on the accumulated history information, the selection of the collation information for use by the speech recognition collating means and/or selection of the correction information for use by the sentence correction means. Thus, since the AI (artificial intelligence) cumulative learning function is used to select collation information for the speech translation and correction information for the correction of translated sentence, the speech translation can be done at an improved speed and with an improved performance and can also be optimized for each user.
Also the above object can be attained by providing a speech translating method including the steps of:
extracting features of a supplied speech to generate a feature vector;
collating the generated feature vector with a plurality of collation information for speech recognition to recognize the speech;
translating the sentence recognized by the speech recognition into a sentence in a language different from that of the supplied speech;
correcting the translated sentence by the use of optimum one of a plurality of correction information;
comparing the speech recognition result with the history information indicative of the tendency of the speech-recognized sentence at each speech recognition to update the history information by a cumulative learning function;
controlling, based on the updated history information, the selection of the collation information used for the speech recognition and/or selection of the correction information used for the sentence correction; and
outputting the sentence corrected by the sentence correction.
In the speech translating method, the speech rec
Frommer William S.
Frommer & Lawrence & Haug LLP
Knepper David D.
Sony Corporation
LandOfFree
Speech translator, speech translating method, and recorded... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Speech translator, speech translating method, and recorded..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Speech translator, speech translating method, and recorded... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3097613