SIMULTANEOUS SPEECH PROCESSING APPARATUS AND METHOD

• US 20150081272A1
• Filed: 09/19/2014
• Published: 03/19/2015
• Est. Priority Date: 09/19/2013
• Status: Active Grant

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation), Toshiba Digital Solutions Corporation (Toshiba Corporation)
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: KAMATANI, Satoshi, SAKAMOTO, Akiko
• Granted Patent: US 9,672,820 B2
• Time in Patent Office: Days
• US Class Current: 704/2
• CPC Class Codes: G06F 40/284 Lexical analysis, e.g. tokenisation or collocates
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence

Abstract

According to one embodiment, a simultaneous speech processing apparatus includes an acquisition unit, a speech recognition unit, a detection unit and an output unit. The acquisition unit acquires a speech signal. The speech recognition unit generates a decided character string and at least one candidate character string. The detection unit detects a first character string as a processing piece character string if the first character string included in the decided character string exists commonly in one or more combined character strings on dividing the one or more combined character strings by a boundary indicating a morphological position serving as a start position of a processing piece in natural language processing. The output unit outputs the processing piece character string.

Claims

1. A simultaneous speech processing apparatus, comprising:

an acquisition unit configured to acquire a speech signal;

a speech recognition unit configured to generate a decided character string and at least one candidate character string, the decided character string being a character string corresponding to a speech section in which part of the speech signal having undergone speech recognition processing is converted into a character string, the at least one candidate character string being a character string corresponding to a speech section in which a character string as a conversion result is variable during speech recognition processing in a speech section succeeding the decided character string;

a detection unit configured to detect a first character string as a processing piece character string if the first character string included in the decided character string exists commonly in one or more combined character strings on dividing the one or more combined character strings by a boundary, the one or more combined character strings being obtained by connecting the decided character string and the at least one candidate character string, the boundary indicating a morphological position serving as a start position of a processing piece in natural language processing; and

an output unit configured to output the processing piece character string.

2. The apparatus according to claim 1, further comprising an update unit configured to update a previously acquired processing piece character string if a position of the boundary changes in accordance with a relationship between a newly acquired processing piece character string and the previously acquired processing piece character string.

3. The apparatus according to claim 1, further comprising:

a storage configured to store, as processing piece information, the processing piece character string and time information of a speech signal corresponding to a speech section in which the processing piece character string is uttered, in association with each other; and

an update unit configured to, if first processing piece information as new processing piece information is added in the storage and second processing piece information exists, connect processing piece character strings included in the second processing piece information and the first processing piece information in time series to generate a reprocessing piece character string, and update the processing piece information stored in the storage with the reprocessing piece character string and time information corresponding to the reprocessing piece character string, the second processing piece information preceding the first processing piece information and corresponding to a speech section in which the second processing piece information is uttered continuously within a time falling within a threshold value.

4. The apparatus according to claim 3, wherein the speech recognition unit acquires time information about a time during which the processing piece character string is uttered, and

the update unit determines with reference to the time information whether or not the second processing piece information exists.

5. The apparatus according to claim 3, wherein

the speech recognition unit acquires a speech rate which indicates a rate at which a speaker speaks, and

the update unit determines with reference to the speech rate whether or not the second processing piece information exists.

6. The apparatus according to claim 1, wherein if the natural language processing is machine translation, the processing piece is a translation unit suitable for simultaneously and parallelly translating the speech signal.

7. The apparatus according to claim 1, wherein if the natural language processing is speech interaction, the processing piece is a unit for simultaneously and parallelly outputting the speech signal as a speech interaction task.

8. A simultaneous speech processing method, comprising;

acquiring a speech signal;

generating a decided character string and at least one candidate character string, the decided character string being a character string corresponding to a speech section in which part of the speech signal having undergone speech recognition processing is converted into a character string, the at least one candidate character string being a character string corresponding to a speech section in which a character string as a conversion result is variable during speech recognition processing in a speech section succeeding the decided character string;

detecting a first character string as a processing piece character string if the first character string included in the decided character string exists commonly in one or more combined character strings on dividing the one or more combined character strings by a boundary, the one or more combined character strings being obtained by connecting the decided character string and the at least one candidate character string, the boundary indicating a morphological position serving as a start position of a processing piece in natural language processing; and

outputting the processing piece character string.

9. The method according to claim 8, further comprising updating a previously acquired processing piece character string if a position of the boundary changes in accordance with a relationship between a newly acquired processing piece character string and the previously acquired processing piece character string.

10. The method according to claim 8, further comprising;

storing, as processing piece information, the processing piece character string and time information of a speech signal corresponding to a speech section in which the processing piece character string is uttered, in association with each other in the storage; and

connecting, if first processing piece information as new processing piece information is added in the storage and second processing piece information exists, processing piece character strings included in the second processing piece information and the first processing piece information in time series to generate a reprocessing piece character string, and updating the processing piece information stored in the storage with the reprocessing piece character string and time information corresponding to the reprocessing piece character string, the second processing piece information preceding the first processing piece information and corresponding to a speech section in which the second processing piece information is uttered continuously within a time falling within a threshold value.

11. The method according to claim 10, wherein the generating the decided character string acquires time information about a time during which the processing piece character string is uttered, and

the updating the processing piece information determines with reference to the time information whether or not the second processing piece information exists.

12. The method according to claim 10, wherein

the generating the decided character string acquires a speech rate which indicates a rate at which a speaker speaks, and

the updating the processing piece information determines with reference to the speech rate whether or not the second processing piece information exists.

13. The method according to claim 8, wherein if the natural language processing is machine translation, the processing piece is a translation unit suitable for simultaneously and parallelly translating the speech signal.

14. The method according to claim 8, wherein if the natural language processing is speech interaction, the processing piece is a unit for simultaneously and parallelly outputting the speech signal as a speech interaction task.

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

acquiring a speech signal;

generating a decided character string and at least one candidate character string, the decided character string being a character string corresponding to a speech section in which part of the speech signal having undergone speech recognition processing is converted into a character string, the at least one candidate character string being a character string corresponding to a speech section in which a character string as a conversion result is variable during speech recognition processing in a speech section succeeding the decided character string;

detecting a first character string as a processing piece character string if the first character string included in the decided character string exists commonly in one or more combined character strings on dividing the one or more combined character strings by a boundary, the one or more combined character strings being obtained by connecting the decided character string and the at least one candidate character string, the boundary indicating a morphological position serving as a start position of a processing piece in natural language processing; and

outputting the processing piece character string.

16. The medium according to claim 15, further comprising updating a previously acquired processing piece character string if a position of the boundary changes in accordance with a relationship between a newly acquired processing piece character string and the previously acquired processing piece character string.

17. The medium according to claim 15, further comprising:

storing, as processing piece information, the processing piece character string and time information of a speech signal corresponding to a speech section in which the processing piece character string is uttered, in association with each other in the storage; and

connecting, if first processing piece information as new processing piece information is added in the storage and second processing piece information exists, processing piece character strings included in the second processing piece information and the first processing piece information in time series to generate a reprocessing piece character string, and updating the processing piece information stored in the storage with the reprocessing piece character string and time information corresponding to the reprocessing piece character string, the second processing piece information preceding the first processing piece information and corresponding to a speech section in which the second processing piece information is uttered continuously within a time falling within a threshold value.

18. The medium according to claim 17, wherein the generating the decided character string acquires time information about a time during which the processing piece character string is uttered, and

the updating the processing piece information determines with reference to the time information whether or not the second processing piece information exists.

19. The medium according to claim 17, wherein

the generating the decided character string acquires a speech rate which indicates a rate at which a speaker speaks, and

the updating the processing piece information determines with reference to the speech rate whether or not the second processing piece information exists.

20. The medium according to claim 15, wherein if the natural language processing is machine translation, the processing piece is a translation unit suitable for simultaneously and parallelly translating the speech signal.