Speech recognition apparatus and speech recognition program

• US 7,818,171 B2
• Filed: 03/13/2007
• Issued: 10/19/2010
• Est. Priority Date: 03/31/2006
• Status: Expired due to Fees

• Current Assignee: DENSO Corporation
• Original Assignee: DENSO Corporation
• Inventors: Hitotsumatsu, Takafumi, Takami, Masayuki
• Primary Examiner(s): ALBERTALLI, BRIAN LOUIS
• Application Number:
  US11/717,027
• Publication Number:
  US 20070233485A1
• Time in Patent Office: 1,316 Days
• Field of Search: None
• US Class Current: 704/252
• CPC Class Codes: G10L 15/08 Speech classification or search

Abstract

A speech recognition apparatus includes a storage medium for storing tree structured dictionary data containing words as nodes in a tree structure with a root and leaf nodes. An input speech is compared with a forward acoustic model corresponding to a speech resulting from chronologically reproducing words indicated by nodes for leaf nodes. A backward speech is further compared with a backward acoustic model. The backward speech is generated by reproducing the input speech in chronologically backward order. The backward acoustic model corresponds to a speech resulting from reproducing a word string toward the root in chronologically backward order. The comparison is performed in the backward order of a sequence starting from one of separator nodes. The speech recognition apparatus thereby outputs a word string that highly likely matches the input speech.

Claims

1. A speech recognition apparatus comprising:

a storage medium for storing tree structured dictionary data that contains a plurality of words as nodes in a tree structure;

a backward speech comparison unit for comparing a backward speech resulting from reproducing an input speech in chronologically backward order with a backward acoustic model corresponding to a speech of reversely reproducing a word string sequence toward a root of the tree structure, wherein a comparison is performed in reverse order of the sequence;

a forward speech comparison unit for comparing the input speech with a forward acoustic model corresponding to a speech resulting from reproducing a leaf node word of the tree structure in chronologically forward order;

a detection unit for detecting a beginning and an end of receiving the input speech, wherein the forward speech comparison unit starts comparing received part of the input speech with the forward acoustic model immediately when the detection unit detects a beginning of receiving the input speech before detecting an end of receiving the input speech; and

an output unit for outputting a word or a word string highly likely matching the input speech based on a comparison result each from the backward speech comparison unit and the forward speech comparison unit.

2. The speech recognition apparatus of claim 1,

wherein the storage medium stores leaf node dictionary data containing a word same as a leaf node word according to the tree structure in the tree structured dictionary data;

and

wherein the forward speech comparison unit compares the input speech with a forward acoustic model corresponding to a speech resulting from chronologically reproducing the word recorded in the leaf node dictionary data.

3. The speech recognition apparatus of claim 1,

wherein the storage medium stores leaf specification data for specifying a leaf node word according to the tree structure in the tree structured dictionary data;

and

wherein the forward speech comparison unit compares the input speech with a forward acoustic model corresponding to a speech resulting from chronologically reproducing the word specified by the leaf specification data.

4. The speech recognition apparatus of claim 1, comprising:

a selection unit for selecting a plurality of words from a group of leaf node words according to the tree structure,

wherein the forward speech comparison unit compares the input speech with a forward acoustic model corresponding to a speech resulting from chronologically reproducing the plurality of words selected by the selection unit.

5. The speech recognition apparatus of claim 4, comprising:

a place name specification unit for specifying an address name corresponding to a current position of the speech recognition apparatus,

wherein the tree structured dictionary data contains a sequence of words from a root to a leaf node of a tree structure, and each word represents one address name, and

wherein the selection unit selects a word from a group of the leaf node words based on the address name specified by the place name specification unit.

6. The speech recognition apparatus of claim 4,

wherein the selection unit selects a word from a group of the leaf node words based on a word or a word string previously output by the output unit.

7. The speech recognition apparatus of claim 6,

wherein the output unit replaces part of a highly likely matching word string previously output by the output unit with a most recent, highly likely matching leaf node word and outputs a word string as a replacement result.

8. A speech recognition apparatus comprising:

a storage medium for storing tree structured dictionary data that contains a plurality of words as nodes in a tree structure;

a backward speech comparison unit for comparing a backward speech resulting from reproducing an input speech in chronologically backward order with a backward acoustic model corresponding to a speech of reproducing, in chronologically backward order, a word string sequence toward a root of the tree structure, wherein a comparison is performed in reverse order of the sequence;

a forward speech comparison unit for comparing the input speech with a forward acoustic model corresponding to a speech resulting from chronologically reproducing an intermediate word string, which is a sequence of a word string starting from a leaf node word toward a root and ending with a node other than the root of the tree structure, wherein a comparison is performed in order of the sequence;

a detection unit for detecting a beginning and an end of receiving the input speech, wherein the forward speech comparison unit starts comparing a received part of the input speech with the forward acoustic model immediately when the detection unit detects a beginning of receiving the input speech before detecting an end of receiving the input speech; and

an output unit for outputting a word string highly likely matching the input speech based on a comparison result each from the backward speech comparison unit and the forward speech comparison unit.

9. The speech recognition apparatus of claim 8,

wherein the storage medium stores leaf node dictionary data containing a word string same as the intermediate word string;

and

wherein the forward speech comparison unit compares the input speech with a forward acoustic model corresponding to a speech resulting from chronologically reproducing the word string recorded in the leaf node dictionary data in order of a sequence of the word string.

10. The speech recognition apparatus of claim 8,

wherein the storage medium stores leaf specification data for specifying the intermediate word string in the tree structured dictionary data;

and

wherein the forward speech comparison unit compares the input speech with a forward acoustic model corresponding to a speech resulting from chronologically reproducing the intermediate word string specified by the leaf specification data in order of the sequence of the intermediate word string.

11. The speech recognition apparatus of claim 8, comprising:

a selection unit for selecting the word string from a sequence of word strings that starts from a leaf node word toward a root and ends with a node other than the root of the tree structure,

wherein the forward speech comparison unit compares, in order of the sequence, the input speech with a forward acoustic model corresponding to a speech resulting from chronologically reproducing the intermediate word string selected by the selection unit.

12. The speech recognition apparatus of claim 11, comprising:

a place name specification unit for specifying an address name corresponding to a current position of the speech recognition apparatus,

wherein the tree structured dictionary data contains a sequence of words from a root to a leaf node of a tree structure, and each word represents one address name, and

wherein the selection unit selects the intermediate word string based on the address name specified by the place name specification unit.

13. The speech recognition apparatus of claim 11,

wherein the selection unit selects the intermediate word string based on a highly likely matching word string previously output by the output unit.

14. The speech recognition apparatus of claim 13,

wherein the output unit replaces part of a highly likely matching word string previously output by the output unit with a most recent, highly likely matching intermediate word string and outputs a word string as a replacement result.

15. A speech recognition apparatus comprising:

a storage medium for storing tree structured dictionary data and separator specification data, wherein the tree structured dictionary data contains a plurality of words as nodes in a tree structure and the separator specification data specifies a plurality of nodes equivalent to separators for words of the tree structured dictionary data other than a root in the tree structure;

a backward speech comparison unit for comparing a backward speech resulting from reproducing an input speech in chronologically backward order with a backward acoustic model corresponding to a speech of reproducing, in chronologically backward order, a sequence of words or word string that is arranged toward a root of the tree structure and ends with one of the plurality of nodes specified by the separator specification data, wherein a comparison is performed in reverse order of the sequence; and

an output unit for outputting a word or a word string highly likely matching the input speech based on a comparison result from the backward speech comparison unit.

16. The speech recognition apparatus of claim 15, comprising:

a selection unit for selecting part of a plurality of nodes specified by the plurality of pieces of separator specification data,

wherein the backward speech comparison unit compares the backward speech with a backward acoustic model corresponding to a speech resulting from reproducing, in chronologically backward order, a sequence of words or word string ending with the selected node; and

wherein a comparison is performed in reverse order of the sequence.

17. The speech recognition apparatus of claim 15, comprising:

a forward speech comparison unit for comparing the input speech with a forward acoustic model corresponding to a speech resulting from chronologically reproducing a sequence of words that starts from a leaf node word toward a root and ends with a node other than the root of the tree structure, wherein a comparison is performed in order of the sequence,

wherein the node specified by the separator specification data corresponds to a node other than a separator for a word nearest to a leaf node of the tree structure in the tree structured dictionary data; and

wherein the output unit outputs a word string highly likely matching the input speech based on a comparison result each from the backward speech comparison unit and the forward speech comparison unit.

18. The speech recognition apparatus of claim 15, comprising:

a forward speech comparison unit for comparing the input speech with a forward acoustic model corresponding to a speech resulting from reproducing a leaf node word of the tree structure in chronologically forward order,

wherein the node specified by the separator specification data corresponds to a node other than a separator for a word nearest to a leaf node of the tree structure in the tree structured dictionary data; and

wherein the output unit outputs a word or a word string highly likely matching the input speech based on a comparison result each from the backward speech comparison unit and the forward speech comparison unit.

19. A computer program product in a computer-readable medium for use in recognizing a speech, the product comprising:

instructions for performing backward speech comparison to compare a backward speech resulting from reproducing an input speech in chronologically backward order with a backward acoustic model corresponding to a speech of reversely reproducing a word string sequence toward a root of a tree structure for tree structured dictionary data containing a plurality of words as nodes in the tree structure, wherein the comparison is made in reverse order of the sequence;

instructions for performing forward speech comparison to compare the input speech with a forward acoustic model corresponding to a speech resulting from reproducing a leaf node word of the tree structure in chronologically forward order, wherein comparing of a received part of the input speech with the forward acoustic model is started immediately when a beginning of receiving the input speech is detected, before an end of receiving the input speech is detected; and

instructions for outputting a word or a word string highly likely matching the input speech based on a comparison result each from the backward speech comparison and the forward speech comparison.

20. A computer program product in a computer-readable medium for use in recognizing a speech, the product comprising:

instructions for performing backward speech comparison to compare a backward speech resulting from reproducing an input speech in chronologically backward order with a backward acoustic model corresponding to a speech of reproducing, in chronologically backward order, a word string sequence toward a root of a tree structure for tree structured dictionary data containing a plurality of words as nodes in the tree structure, wherein the comparison is performed in reverse order of the sequence;

instructions for performing forward speech to compare the input speech with a forward acoustic model corresponding to a speech resulting from chronologically reproducing an intermediate word string, which is a sequence of a word string starting from a leaf node word toward a root and ending with a node other than the root of the tree structure, wherein the comparison is performed in order of the sequence, wherein comparing of a received part of the input speech with the forward acoustic model is started immediately when a beginning of receiving the input speech is detected, before an end of receiving the input speech is detected; and

instructions for outputting a word string highly likely matching the input speech based on a comparison result each from the backward speech comparison and the forward speech comparison.

21. A computer program product in a computer-readable medium for use in recognizing a speech, the product comprising:

instructions for reading tree structured dictionary data and separator specification data from a storage medium for storing the data, wherein the tree structured dictionary data contains a plurality of words as nodes in a tree structure and the separator specification data specifies a plurality of nodes equivalent to separators for words of the tree structured dictionary data other than a root in the tree structure;

instructions for performing backward speech comparison to compare a backward speech resulting from reproducing an input speech in chronologically backward order with a backward acoustic model corresponding to a speech of reproducing, in chronologically backward order, a sequence of words or word string that is arranged toward a root of the tree structure and ends with one of the plurality of nodes specified by the separator specification data, wherein the comparison is performed in reverse order of the sequence; and

instructions for outputting a word or a word string highly likely matching the input speech based on a comparison result from the backward speech comparison.