Difference extraction device, method and program

1. A difference extraction device comprising processing circuitry configured to:

acquire a text in which an input notation string is described;

convert the input notation string into a pronunciation string;

convert the pronunciation string into an output notation string;

extract a difference by comparing the input notation string and the output notation string with each other, and determining, as the difference, based on a result of the comparison, a part of the output notation string which differs from a part of the input notation string, wherein said parts are respectively located at corresponding positions in the input notation string and the output notation string;

analyze the input notation string;

estimate a notation of a word candidate contained in the input notation string and including the difference, based on an analysis result of the input notation string; and

determine a word type of the word candidate,

wherein:

where the notation of the word candidate is estimated, the processing circuitry extracts, from the input notation string, a character string estimated to form a word by concatenating the difference and at least one character adjacent to the difference, and outputs the extracted character string as the estimated notation of the word candidate,

where the word type of the word candidate is determined, the processing circuitry determines, using a different notation dictionary which is a dictionary listing different notations of a same word with a same meaning, if two notations including the notation of the word candidate and the notation in the output notation string corresponding to the notation of the word candidate are present in the different notation dictionary, that the two notations are caused by a notation fluctuation,

where the pronunciation string is converted into the output notation string,

the processing circuitry is further configured to:

convert the pronunciation string into an acoustic score vector;

store a language model and a word dictionary for speech recognition in a memory; and

generate a second pronunciation string from the acoustic score vector and convert the generated second pronunciation string into the output notation string, using the language model and the word dictionary,

wherein the acoustic score vector is a 102-dimensional vector representing a likelihood of each of syllables included in the pronunciation string that is in a Japanese language, a pronunciation of each of the syllables being expressed by a single state, the Japanese language including 102 syllables, and

wherein the processing circuitry is further configured to execute a display control process in which the input notation string including the word candidate is displayed on a display and in which the notation of the word candidate is displayed on the display, using a display attribute based on the word candidate.

2. The difference extraction device according to claim 1 , wherein, in the display control process, the processing circuitry is further configured to:

indicate a range of a notation contained in the displayed input notation string and including at least part of the word candidate.

3. The difference extraction device according to claim 1 , wherein in the display control process, the notation of the word candidate is displayed on the display using the display attribute, the display attribute corresponding to the word type.

4. The difference extraction device according to claim 2 , wherein the processing circuit is further configured to register the notation of the indicated range in a word dictionary.

5. The difference extraction device according to claim 1 , wherein

where the notation of the word candidate is not registered in a word dictionary, the processing circuitry is further configured to determine that the notation of the word candidate is an unknown word.

6. The difference extraction device according to claim 1 , wherein

where the pronunciation string is converted into the acoustic score vector, the processing circuitry is further configured to synthesize an audio signal from the pronunciation string, calculate an acoustic feature vector from the audio signal, and calculate an acoustic score vector from the acoustic feature vector.

7. The difference extraction device according to claim 1 , wherein

where the input notation string is converted into the pronunciation string, the processing circuitry is further configured to divide the input notation string into words, perform a morphological analysis for estimating part of speech of each word, and convert the words into the pronunciation string, with readings attached thereto, based on a result of the morphological analysis.

8. A difference extraction method comprising:

acquiring a text in which an input notation string is described;

converting the input notation string into a pronunciation string;

converting the pronunciation string into an output notation string;

extracting a difference by comparing the input notation string and the output notation string with each other, and determining, as the difference, based on a result of the comparison, a part of the output notation string which differs from a part of the input notation string, wherein said parts are respectively located at corresponding positions in the input notation string and the output notation string;

analyzing the input notation string;

estimating a notation of a word candidate contained in the input notation string and including the difference, based on an analysis result of the input notation string; and

determining a word type of the word candidate,

wherein:

where the notation of the word candidate is estimated, the method further comprises extracting, from the input notation string, a character string estimated to form a word by concatenating the difference and at least one character adjacent to the difference, and outputting the extracted character string as the estimated notation of the word candidate,

where the word type of the word candidate is determined, the method further comprises determining, using a different notation dictionary which is a dictionary listing different notations of a same word with a same meaning, if two notations including the notation of the word candidate and the notation in the output notation string corresponding to the notation of the word candidate are present in the different notation dictionary, that the two notations are caused by a notation fluctuation, where the pronunciation string is converted into the output notation string, the method further comprises:

converting the pronunciation string into an acoustic score vector;

storing a language model and a word dictionary for speech recognition in a memory; and

generating a second pronunciation string from the acoustic score vector and converting the generated second pronunciation string into the output notation string, using the language model and the word dictionary,

wherein the acoustic score vector is a 102-dimensional vector representing a likelihood of each of syllables included in the pronunciation string that is in a Japanese language, a pronunciation of each of the syllables being expressed by a single state, the Japanese language including 102 syllables, and

wherein the method further comprises executing a display control process in which the input notation string including the word candidate is displayed on a display and in which the notation of the word candidate is displayed on the display, using a display attribute based on the word candidate.

9. A non-transitory computer readable storage medium including computer executable instructions, wherein the instructions, when executed by a processor, cause the processor to perform a method comprising:

acquiring a text in which an input notation string is described;

converting the input notation string into a pronunciation string;

converting the pronunciation string into an output notation string; and

extracting a difference by comparing the input notation string and the output notation string with each other, and determining, as the difference, based on a result of the comparison, a part of the output notation string which differs from a part of the input notation string, wherein said parts are respectively located at corresponding positions in the input notation string and the output notation string;

analyzing the input notation string;

estimating a notation of a word candidate contained in the input notation string and including the difference, based on an analysis result of the input notation string; and

determining a word type of the word candidate,

wherein:

where the notation of the word candidate is estimated, the method further comprises extracting, from the input notation string, a character string estimated to form a word by concatenating the difference and at least one character adjacent to the difference, and outputting the extracted character string as the estimated notation of the word candidate,

where the word type of the word candidate is determined, the method further comprises determining, using a different notation dictionary which is a dictionary listing different notations of a same word with a same meaning, if two notations including the notation of the word candidate and the notation in the output notation string corresponding to the notation of the word candidate are present in the different notation dictionary, that the two notations are caused by a notation fluctuation, where the pronunciation string is converted into the output notation string, the method further comprises:

converting the pronunciation string into an acoustic score vector;

storing a language model and a word dictionary for speech recognition in a memory; and

generating a second pronunciation string from the acoustic score vector and converting the generated second pronunciation string into the output notation string, using the language model and the word dictionary,

wherein the acoustic score vector is a 102-dimensional vector representing a likelihood of each of syllables included in the pronunciation string that is in a Japanese language, a pronunciation of each of the syllables being expressed by a single state, the Japanese language including 102 syllables, and

wherein the method further comprises executing a display control process in which the input notation string including the word candidate is displayed on a display and in which the notation of the word candidate is displayed on the display, using a display attribute based on the word candidate.