# US 2015/0088506 A1 · Speech Recognition Server Integration Device and Speech Recognition Server Integration Method

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2015/0088506 A1
- Document ID: US20150088506A1
- Status: Published application
- Kind: A1
- Application number: 14/391,200
- Filed: 2013-04-03
- Published: 2015-03-26
- Earliest claimed: 2012-04-09
- Assignees: Clarion Co., Ltd.
- Applicants: Clarion Co., Ltd.
- Inventors: Yasunari Obuchi; Takeshi Homma
- Web page: https://patentgrep.com/patent/US20150088506A1

## CPC

- G10L 15/30: Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
- G10L 15/08: Speech classification or search
- G10L 15/32: Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems
- G10L 15/22: Procedures used during a speech recognition process, e.g. man-machine dialogue
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

The speech recognition result through the general-purpose server and that through the specialized speech recognition server are integrated in an optimum manner, thereby, a speech recognition function least in errors in the end being provided. The specialized speech recognition server 108 is constructed with the words contained in the user dictionary data in use as well as the performance of the general-purpose speech recognition server 106 is preliminarily evaluated with such user dictionary data. Based on such evaluation result, information related to which recognition results through the specialized and general-purpose speech recognition servers are adopted and to how the adopted recognition results are weighted to obtain an optimum recognition result is preliminarily retained in the form of a database. Upon executing recognition, an optimum recognition result is obtained by comparing the recognition results through the specialized and general-purpose servers with the parameter for recognition result integration 118.

## Claims

### Claim 1

A speech recognition server integration device, which is an intermediate module to relay between a terminal module by which a user performs operations by way of speech and a speech recognition server to recognize speech data and to feed back its recognition result, comprising:
an integration-manner learning module to learn and preserve a parameter for recognition result integration based on words registered by the user or a list of words frequently used by the user;
means to receive speech data spoken by the user for speech recognition from the terminal module;
means to transmit the received speech data to a general-purpose speech recognition server and a specialized speech recognition server;
means to receive recognition results of the speech data through the general-purpose speech recognition server and the specialized speech recognition server;
a recognition result integration module to compare the recognition results through the general-purpose speech recognition server and the specialized speech recognition server with the preserved parameter for recognition result integration and to select an optimum recognition result; and
means to transmit the selected recognition result to the terminal module.

### Claim 2 (depends on 1)

The speech recognition server integration device according to claim 1, further comprising:
means to receive the words registered by the user or the list of words frequently used by the user from the terminal module;
a module of making synthesized speech based on the received words;
means to transmit the made synthesized speech to the general-purpose speech recognition server and the specialized speech recognition server; and
means to receive the recognition results of the synthesized speech through the general-purpose speech recognition server and the specialized speech recognition server,
wherein the integration-manner learning module concurrently analyses the words from which the synthesized speech derives and the recognition results so as to learn and preserve the parameter for recognition result integration.

### Claim 3 (depends on 1)

The speech recognition server integration device according to claim 1, further comprising:
means to receive the words registered by the user or the list of words frequently used by the user from the terminal module;
means to receive a list of words for recognition from the general-purpose speech recognition server; and
a words comparison/similarity estimation module to compare the list of words for recognition with the list of words received from the terminal module and to estimate similarity between the lists,
wherein the integration-manner learning module preserves an estimation result as the parameter for recognition result integration.

### Claim 4

A speech recognition server integration device, which is a module by which a user performs operations by way of speech, comprising:
an integration-manner learning module to learn and preserve a parameter for recognition result integration based on words registered by the user or a list of words frequently used by the user;
means to transmit data of a speech spoken by the user for speech recognition to a general-purpose speech recognition server and a specialized speech recognition server;
means to receive recognition results of the speech data through the general-purpose speech recognition server and the specialized speech recognition server;
a recognition result integration module to compare the recognition results through the general-purpose speech recognition server and the specialized speech recognition server with the preserved parameter for recognition result integration and to select an optimum recognition result; and
a display module to display the selected recognition result.

### Claim 5 (depends on 4)

The speech recognition server integration device according to claim 4, further comprising:
a user dictionary to store the words registered by the user or the list of words frequently used by the user;
a speech synthesis module to make synthesized speech based on the words stored in the user dictionary;
means to transmit the made synthesized speech to the general-purpose speech recognition server and the specialized speech recognition server; and
means to receive recognition results of the synthesized speech through the general-purpose speech recognition server and the specialized speech recognition server,
wherein the integration-manner learning module concurrently analyses the words from which the synthesized speech derives and the recognition results so as to learn and preserve the parameter for recognition result integration.

### Claim 6 (depends on 4)

The speech recognition server integration device according to claim 4, further comprising:
a user dictionary to store the words registered by the user or the list of words frequently used by the user;
means to receive a list of words for recognition from the general-purpose speech recognition server; and
a words comparison/similarity estimation module to compare the list of words for recognition with the list of words of the user dictionary and to estimate similarity between the lists,
wherein the integration-manner learning module preserves an estimation result as the parameter for recognition result integration.

### Claim 7 (depends on 1)

The speech recognition server integration device according to claim 1,
wherein the specialized speech recognition server makes a list of words for recognition target based on the words registered by the user or the list of words frequently used by the user and is capable of recognizing words contained in the list of words for recognition target with high precision.

### Claim 8 (depends on 1)

The speech recognition server integration device according to claim 1,
wherein the specialized speech recognition server is incorporated in the speech recognition server integration device or the terminal module as a specialized speech recognition module.

### Claim 9 (depends on 2)

The speech recognition server integration device according to claim 2,
wherein the parameter for recognition result integration accumulates accuracies and errors of the recognition results of the words registered or frequently used by the user through the speech recognition servers,
wherein the recognition result integration module extracts the recognition results of a certain word through the speech recognition servers from the parameter for recognition result integration based on the recognition result through the specialized speech recognition server and extracts only the recognition results through the speech recognition servers whose recognition results extracted by the recognition result integration module are accurate so as to select an optimum recognition result based on the extracted recognition results.

### Claim 10 (depends on 2)

The speech recognition server integration device according to claim 2,
wherein the parameter for recognition result integration accumulates accuracies and errors of the recognition results of the words registered and frequently used by the user through the speech recognition servers and values denoting confidence measures of the recognition results of the individual words through the speech recognition servers,
wherein the recognition result integration module extracts the recognition results and the confidence measures of the recognition results of a certain word through the speech recognition servers from the parameter for recognition result integration based on the recognition result through the specialized speech recognition server and extracts only the recognition results and the confidence measures of the recognition results through the speech recognition servers whose recognition results extracted by the recognition result integration module are accurate so as to integrate the extracted recognition results by weighting with the confidence measure on the extracted recognition results.

### Claim 11 (depends on 2)

The speech recognition server integration device according to claim 2,
wherein the parameter for recognition result integration measures time required for the speech recognition servers recognizing the words registered or frequently used by the user and accumulates the measured time values,
wherein the recognition result integration module extracts time required for the speech recognition servers recognizing a certain word from the parameter for recognition result integration based on the recognition result through the specialized speech recognition server and obtains an allowable upper value of the time required for the speech recognition servers recognizing a certain word, which value is determined depending on an application in use, and extracts only the recognition results through the speech recognition servers whose time required for recognizing a certain word goes below the allowable upper value so as to select an optimum recognition result based on the extracted recognition results.

### Claim 12 (depends on 2)

The speech recognition server integration device according to claim 2,
wherein the parameter for recognition result integration accumulates accuracies and errors of the recognition results of the words registered or frequently used by the user and one misrecognition result or a plurality of misrecognition results through the speech recognition servers,
wherein the recognition result integration module extracts the accuracies and errors as well as misrecognition results of the recognition results of a certain word through the speech recognition servers from the parameter for recognition result integration based on the recognition result through the specialized speech recognition server and compares the extracted misrecognition result with a recognition result upon the recognition being executed when the extracted recognition result is an error and makes the recognition result effective only when it is determined that the comparison results are the same so as to select an optimum recognition result based on the recognition result made effective.

### Claim 13

A speech recognition server integration method comprising:
a step of learning and preserving a parameter for recognition result integration based on words registered by a user or a list of words frequently used by the user;
a step to transmit data of a speech spoken by the user for speech recognition to a general-purpose speech recognition server and a specialized speech recognition server;
a step to receive recognition results of the speech data through the general-purpose speech recognition server and the specialized speech recognition server; and
a step to compare the recognition result through the general-purpose speech recognition server and the recognition result through the specialized speech recognition server with the parameter for recognition result integration so as to select an optimum speech recognition result.

### Claim 14 (depends on 13)

The speech recognition server integration method according to claim 13, further comprising:
a step of making a synthesized speech based on the words registered or frequently used by the user;
a step of transmitting the made synthesized speech to the general-purpose speech recognition server and the specialized speech recognition server; and
a step of receiving the recognition results of the synthesized speech through the general-purpose speech recognition server and the specialized speech recognition server,
wherein in the step of learning and preserving the parameter for recognition result integration the words from which the synthesized speech derives and the recognition results are concurrently analyzed so as to make the parameter for recognition result integration learnt and preserved.

### Claim 15 (depends on 13)

The speech recognition server integration method according to claim 13, further comprising:
a step of acquiring the words registered by the user or the list of words frequently used by the user;
a step of receiving a list of words for recognition from the general-purpose speech recognition servers; and
a step of comparing the list of words for recognition with the words registered by the user or the list of words frequently used by the user and estimating similarity between them,
wherein in the step of learning and preserving the parameter for recognition result integration the estimation result is preserved as the parameter for recognition result integration.

### Claim 16 (depends on 5)

The speech recognition server integration device according to claim 5,
wherein the parameter for recognition result integration accumulates accuracies and errors of the recognition results of the words registered or frequently used by the user through the speech recognition servers,
wherein the recognition result integration module extracts the recognition results of a certain word through the speech recognition servers from the parameter for recognition result integration based on the recognition result through the specialized speech recognition server and extracts only the recognition results through the speech recognition servers whose recognition results extracted by the recognition result integration module are accurate so as to select an optimum recognition result based on the extracted recognition results.

### Claim 17 (depends on 5)

The speech recognition server integration device according to claim 5,
wherein the parameter for recognition result integration accumulates accuracies and errors of the recognition results of the words registered and frequently used by the user through the speech recognition servers and values denoting confidence measures of the recognition results of the individual words through the speech recognition servers,
wherein the recognition result integration module extracts the recognition results and the confidence measures of the recognition results of a certain word through the speech recognition servers from the parameter for recognition result integration based on the recognition result through the specialized speech recognition server and extracts only the recognition results and the confidence measures of the recognition results through the speech recognition servers whose recognition results extracted by the recognition result integration module are accurate so as to integrate the extracted recognition results by weighting with the confidence measure on the extracted recognition results.

### Claim 18 (depends on 5)

The speech recognition server integration device according to claim 5,
wherein the parameter for recognition result integration measures time required for the speech recognition servers recognizing the words registered or frequently used by the user and accumulates the measured time values,
wherein the recognition result integration module extracts time required for the speech recognition servers recognizing a certain word from the parameter for recognition result integration based on the recognition result through the specialized speech recognition server and obtains an allowable upper value of the time required for the speech recognition servers recognizing a certain word, which value is determined depending on an application in use, and extracts only the recognition results through the speech recognition servers whose time required for recognizing a certain word goes below the allowable upper value so as to select an optimum recognition result based on the extracted recognition results.

### Claim 19 (depends on 5)

The speech recognition server integration device according to claim 5,
wherein the parameter for recognition result integration accumulates accuracies and errors of the recognition results of the words registered or frequently used by the user and one misrecognition result or a plurality of misrecognition results through the speech recognition servers,
wherein the recognition result integration module extracts the accuracies and errors as well as misrecognition results of the recognition results of a certain word through the speech recognition servers from the parameter for recognition result integration based on the recognition result through the specialized speech recognition server and compares the extracted misrecognition result with a recognition result upon the recognition being executed when the extracted recognition result is an error and makes the recognition result effective only when it is determined that the comparison results are the same so as to select an optimum recognition result based on the recognition result made effective.
