INPUT GENERATION FOR CLASSIFIER

• US 20180218736A1
• Filed: 02/02/2017
• Published: 08/02/2018
• Est. Priority Date: 02/02/2017
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Itoh, Nobuyasu, Kurata, Gakuto, Tachibana, Ryuki
• Granted Patent: US 10,540,963 B2
• Time in Patent Office: Days
• CPC Class Codes: G10L 15/01 Assessment or evaluation of speech recognition systems
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 2015/0631 Creating reference templates; Clustering

Abstract

A computer-implemented method for generating an input for a classifier. The method includes obtaining n-best hypotheses which is an output of an automatic speech recognition (ASR) for an utterance, combining the n-best hypotheses horizontally in a predetermined order with a separator between each pair of hypotheses, and outputting the combined n-best hypotheses as a single text input to a classifier.

Claims

1. A computer-implemented method for improving performance of a speech recognition system, comprising:

obtaining n-best hypotheses as an output of a speech recognition task performed by automatic speech recognition (ASR) for an utterance received by the speech recognition system;

combining the n-best hypotheses horizontally in a predetermined order with a separator between each pair of hypotheses to provide combined n-best hypotheses represented as a single text data structure, wherein each separator is set based on a classification algorithm of a classifier of the speech recognition system; and

outputting the single text data structure as an input to the classifier.

3. The method according to claim 1, wherein each of the n-best hypotheses is associated with a corresponding confidence score, and wherein the predetermined order is determined based on the confidence scores.

4. The method according to claim 1, further comprising:

obtaining source data;

generating a plurality of inputs from the source data as training data for the classifier; and

outputting each input from the plurality of inputs as the single text data structure to the classifier.

5. The method according to claim 4, wherein the source data includes a plurality of sets of n-best hypotheses and a transcription for each utterance, and generating the plurality of inputs includes arranging the n-best hypotheses in the predetermined order with the transcription at a head or at an end of the combined n-best hypotheses depending on the predetermined order for each set.

6. The method according to claim 4, wherein the source data includes a plurality of sets of in text samples belonging to a same class and generating the plurality of inputs includes arranging m text samples based on a degree of similarity for each set.

7. The method according to claim 6, wherein n is any number of hypotheses obtained from the output of the speech recognition task and m is a maximum number of hypotheses obtained from output of the speech recognition task.

8. The method according to claim 6, wherein each similarity is determined by measuring a distance between a first selected text sample and each text sample from among plurality of text samples belonging to the same class as the first selected text sample.

9. The method according to claim 8, wherein each text sample is vectorized by a technique selected from a group including a frequency of word, Term Frequency-Inverse Document Frequency, Latent Semantic Indexing and word to vector representation.

2. (canceled)

10. An apparatus comprising:

a processor; and

one or more non-transitory computer readable storage mediums collectively including instructions that, when executed by the processor, cause the processor to perform a method for improving performance of a speech recognition system, the method comprising;

obtaining n-best hypotheses as an output of a speech recognition task performed by automatic speech recognition (ASR) for an utterance received by the speech recognition system;

combining the n-best hypotheses horizontally in a predetermined order with a separator between each pair of hypotheses to provide combined n-best hypotheses represented as a single text data structure, wherein each separator is set based on a classification algorithm of a classifier of the speech recognition system; and

outputting the single text data structure as an input to the classifier.

12. The apparatus according to claim 10, wherein each of the n-best hypotheses is associated with a corresponding confidence score, and wherein the predetermined order is determined based on the confidence scores.

13. The apparatus according to claim 10, wherein the method further comprises:

obtaining a plurality of sets of n-best hypotheses and a transcription for each utterance;

generating training data by arranging, for each set, the n-best hypotheses in the predetermined order with the transcription at a head or at an end of the combined n-best hypotheses depending on the predetermined order; and

outputting the training data as the single text data structure to the classifier.

14. The apparatus according to claim 10, wherein the method further comprises:

obtaining a plurality of sets of m text samples belonging to a same class;

generating training data by arranging, for each set, m text samples based on a degree of similarity; and

outputting the training data as the single text data structure to the classifier.

15. The apparatus according to claim 14, wherein n is any number of hypotheses obtained from the output of the speech recognition task and m is a maximum number of hypotheses obtained from output of the speech recognition task.

11. (canceled)

16. A computer program product, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, wherein the computer readable storage medium is not a transitory signal per se, the program instructions executable by a computer to cause the computer to perform a method for improving performance of a speech recognition system, the method comprising:

obtaining n-best hypotheses as an output of a speech recognition task performed by automatic speech recognition (ASK) for an utterance received by a speech recognition system;

combining the n-best hypotheses horizontally in a predetermined order with a separator between each pair of hypotheses to provide combined n-best hypotheses represented as a single text data structure, wherein each separator is set based on a classification algorithm of a classifier of the speech recognition system; and

outputting the single text data structure as an input to the classifier.

18. The computer program product according to claim 16, wherein each of the n-best hypotheses is associated with a corresponding confidence score, and wherein the predetermined order is determined based on the confidence scores.

19. The computer program product according to claim 16, wherein the method further comprises:

obtaining a plurality of sets of n-best hypotheses and a transcription for each utterance;

generating training data by arranging, for each set, the n-best hypotheses in the predetermined order with the transcription at a head or at an end of the combined n-best hypotheses depending on the predetermined order; and

outputting the training data as the single text data structure to the classifier.

20. The computer program product according to claim 16, wherein the method further comprises:

obtaining a plurality of sets of m text samples belonging to a same class;

generating training data by arranging, for each set, m text samples based on a degree of similarity; and

outputting the training data as the single text data structure to the classifier.

17. (canceled)