Data processing device, information storage medium storing computer program therefor and data processing method

• US 9,047,562 B2
• Filed: 12/02/2010
• Issued: 06/02/2015
• Est. Priority Date: 01/06/2010
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Okabe, Koji, Hanazawa, Ken, Osada, Seiya, Arakawa, Takayuki, Tanaka, Daisuke
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): Wu, Fuming
• Application Number:
  US13/520,728
• Publication Number:
  US 20120310866A1
• Time in Patent Office: 1,643 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 2015/085 Methods for reducing search complexity, pruning
• CPC Class Codes: G10L 25/24 the extracted parameters being the cepstrum

Abstract

A plurality of pruning measures (PM) are calculated from a feature amount (CV) of test data (TD) which is input, a plurality of isopycnic surfaces (EC) are plotted and set on a threshold space (SS), a threshold curved surface (SC) in which a decrease in at least one of a plurality of pruning measures (PM) causes an increase in at least one thereof is generated using a portion of one isopycnic surface (EC) as a part, a hypothesis curved surface (HC) of subject data (CD) is generated on the threshold space (SS) to set a position intersecting the threshold curved surface (SC) to a pruning threshold (PS), and a plurality of hypotheses of the subject data (CD) are pruned. Thereby, there is provided a data processing device of which at least one of the recognition speed and the recognition accuracy is higher than in the related art.

Claims

1. A data processing device comprising:

a data input unit configured to input test data in which a correct hypothesis is established in a learning mode for each of predetermined inputs, and inputs subject data for a hypothesis search in a search mode for each of the inputs;

a feature amount extraction unit, implemented by at least one hardware processor, which is configured to analyze the input test data and the input subject data and extracts each feature amount;

a hypothesis measure calculation unit configured to calculate a plurality of pruning measures for each of a plurality of hypotheses of the test data and the subject data using the extracted feature amount;

a data plotting unit configured to plot the plurality of hypotheses of the input test data on a threshold space specified by the plurality of pruning measures, corresponding to each of the calculated pruning measures;

an isopycnic surface setting unit configured to set a plurality of isopycnic surfaces on the threshold space, corresponding to densities of the plurality of plotted hypotheses;

a threshold curved surface generation unit configured to generate, on the threshold space, a threshold curved surface made of the plurality of pruning thresholds in which a decrease in at least one of the plurality of pruning measures causes an increase in at least one thereof, using a portion of one surface selected from the plurality of isopycnic surfaces as a part;

a hypothesis curved surface generation unit configured to generate a hypothesis curved surface made of the plurality of hypotheses of the subject data on the threshold space, corresponding to each of the calculated pruning measures; and

a hypothesis pruning unit that configured to prune the plurality of hypotheses of the subject data, using a position at which the generated hypothesis curved surface intersects the threshold curved surface as the pruning threshold.

2. The data processing device according to claim 1, wherein the threshold space is made of a two-dimensional threshold plane surface specified by a score difference of the hypotheses from a maximum likelihood hypothesis and a hypothesis rank which are the two pruning measures,

the hypothesis measure calculation unit is configured to calculate a score for each of the plurality of hypotheses of the test data and the subject data using the extracted feature amount and calculates the score difference and the hypothesis rank as the pruning measures,

the isopycnic surface setting unit is configured to set isopycnic lines which are the plurality of isopycnic surfaces on the threshold plane surface, corresponding to the densities of the plurality of plotted hypotheses,

the threshold curved surface generation unit is configured to generate a threshold curve as the threshold curved surface on the threshold plane surface, the threshold curve being made of the plurality of pruning thresholds in which a decrease in one of the score difference and the hypothesis rank causes an increase in the other thereof, using a portion of one line selected from the plurality of isopycnic lines as a part, and

the hypothesis curved surface generation unit is configured to generate a hypothesis curve made of the plurality of hypotheses of the subject data as the hypothesis curved surface on the threshold plane surface, corresponding to the score difference and the hypothesis rank which are calculated.

3. The data processing device according to claim 2, further comprising a result output unit that is configured to output one hypothesis, having a maximum value of the scores accumulated from the plurality of hypotheses pruned in the search mode, as a search result.

4. The data processing device according to claim 1, wherein the test data and subject data are audio data, and wherein the data processing device is configured to perform audio recognition.

5. The data processing device according to claim 1, wherein the test data and subject data are image data, and wherein the data processing device is configured to perform image recognition.

6. An non-transitory information storage medium storing a computer program, the program causing a data processing device to execute:

a data input process of inputting test data in which a correct hypothesis is established in a learning mode for each of predetermined inputs, and inputting subject data for a hypothesis search in a search mode for each of the predetermined inputs;

a feature amount extraction process of analyzing the input test data and the subject data and extracting each feature amount;

a hypothesis measure calculation process of calculating a plurality of pruning measures for each of a plurality of hypotheses of the test data and the subject data using the extracted feature amount;

a data plotting process of plotting the plurality of hypotheses of the input test data on a threshold space specified by the plurality of pruning measures, corresponding to each of the calculated pruning measures;

an isopycnic surface setting process of setting a plurality of isopycnic surfaces on the threshold space, corresponding to densities of the plurality of plotted hypotheses;

a threshold curved surface generation process of generating, on the threshold space, a threshold curved surface made of the plurality of pruning thresholds in which a decrease in at least one of the plurality of pruning measures causes an increase in at least one thereof, using a portion of one surface selected from the plurality of isopycnic surfaces as a part;

a hypothesis curved surface generation process of generating a hypothesis curved surface made of the plurality of hypotheses of the subject data on the threshold space, corresponding to each of the calculated pruning measures; and

a hypothesis pruning process of pruning the plurality of hypotheses of the subject data, using a position at which the generated hypothesis curved surface intersects the threshold curved surface as the pruning threshold.

7. A data processing method, comprising:

a data input operation of inputting test data in which a correct hypothesis is established in a learning mode for each of predetermined inputs, and inputting subject data for a hypothesis search in a search mode for each of the inputs;

a feature amount extraction operation, implemented by at least one hardware processor, of analyzing the input test data and the subject data and extracting each feature amount;

a hypothesis measure calculation operation of calculating a plurality of pruning measures for each of a plurality of hypotheses of the test data and the subject data using the extracted feature amount;

a data plotting operation of plotting the plurality of hypotheses of the input test data on a threshold space specified by the plurality of pruning measures, corresponding to each of the calculated pruning measures;

an isopycnic surface setting operation of setting a plurality of isopycnic surfaces on the threshold space, corresponding to densities of the plurality of plotted hypotheses;

a threshold curved surface generation operation of generating, on the threshold space, a threshold curved surface made of the plurality of pruning thresholds in which a decrease in at least one of the plurality of pruning measures causes an increase in at least one thereof, using a portion of one surface selected from the plurality of isopycnic surfaces as a part;

a hypothesis curved surface generation operation of generating a hypothesis curved surface made of the plurality of hypotheses of the subject data on the threshold space, corresponding to each of the calculated pruning measures; and

a hypothesis pruning operation of pruning the plurality of hypotheses of the subject data, using a position at which the generated hypothesis curved surface intersects the threshold curved surface as the pruning threshold.