KR20060009151A - Incantation speech recongnition wireless site navigation system & service - Google Patents

Abstract

It is very inconvenient to search for wireless Internet sites / content via mobile phones. This is because the wireless Internet / content address (url) must be entered in several steps using the keypad buttons in the wireless search box or through text-based wireless navigation. This paper proposes a voice navigation system and service through a voice recognition multi-modal platform that goes directly through voice commands. The key idea behind this service, however, is that the voice commands form a unique incantation that represents each site / content, acting as a brand for that wireless site / content and inducing fun, and unlike regular words, the syllable uniqueness of the order. Differentiation can also increase speech recognition rates. For example, if you want to access a wireless site / content, you can say "alohamora" (an order to open the doors of Harry Potter and the Sorcerer's Stone) on your mobile phone. .

To implement this, a distributed speech recognition multimodal platform such as the following representation should be utilized. In other words, when a mobile phone terminal application records a voice order (incantation) and sends the data to a voice recognition multimodal platform, the platform recognizes it and connects it to the corresponding wireless site.

Spelling, incantation, mobile phone applications, speech recognition, voice navigation, distributed speech recognition (DSR), automatic speech recognition (ASR), text-to-speech (TTS)

Description

Incantation speech recongnition wireless site navigation system & service

Figure 1. Configuration of the entire system

Figure 2. Type of DSR

Figure 3. ASR Server

Figure 4. WA server

Figure 5. DSR API

The present invention is a technology that utilizes distributed voice recognition technology to a mobile phone, instead of going to a wireless site / content desired by pressing a keypad button, the user directly speaks the voice command corresponding to the content to the mobile phone and uses the service. . Conventional keypad-based navigation has the inconvenience of inputting through several steps or pressing a direct button. Also, the conventional navigation technology using voice recognition is limited to speech-based content because a voice mode must be used as a single mode. It was not suitable for data site based wireless site / content search. In order to compensate for this, a technique of attaching a voice engine to a terminal or recording a terminal to a voice server has been proposed. As a solution to this problem, we first propose a platform to apply distributed speech recognition technology to a multi-modal base that supports voice and data at the same time. We would like to propose a service model that enhances the fun of the company.

First, based on a multimodal platform in the form of voice input / data output to navigate data-based wireless sites / content. As shown in FIG. 1, when a voice is recorded and transmitted to a mobile phone terminal, the distributed voice recognition engine is preprocessed and sent to an ASR server equipped with a high performance voice recognition algorithm. This process can provide voice / data through the mobile phone screen and can also improve the performance of voice recognition. This proposal proposes a new type of voice navigation service based on such a platform, where voice commands are not simply general words or names (e.g. news, securities, 'Yahoo' and 'green pepper mobile go-stop'). To navigate using) command. For example, if you say "Aloha Mora" (an order used to open the doors of Harry Potter and the Sorcerer's Stone) to access the Bell Pepper Mobile GoStop service, you will be immediately connected to the Bell Pepper Mobile GoStop service site registered with this order. These orders are created by each service provider and registered with Grammar on this multi-modal voice recognition platform, and given this voice command, they are directed to the URL of the site registered with that command.

The configuration of the entire system of the present invention is as follows. See Figure 1

The terminal has a DSR API to communicate with Backend's DSR server, and the DSR server supports services in the Contents server by interworking with ASR server, TTS server, and VXML interpreter server.

end. System configuration

1. DSR

2 shows a form of the DSR of the present system. This means that voice output and screen output are simultaneously performed through interworking between VoiceXML and a terminal application. On the other hand, since it is possible to perform voice input / key input at the terminal using a data channel at the same time, it can be said that it supports DSR-based multimodal UI. In addition, organic interworking with the terminal is possible. That is, the performance of the voice engine can be improved, and tuning is easy without changing the terminal H / W by using the voice call codec (EVRC).

2. ARS Server

ASR server implements and operates the service of voice recognition engine provided in library form. Implemented in the form of a server over TCP, an application program that requires a voice recognition service becomes a client and uses the services of ASR.

The ASR server uses more than one TCP port to provide services to clients and allows multiple clients to simultaneously use the service by allowing multiple TCP socket connections.

The ASR server provides a voice recognition service through a virtual voice recognition channel. Channel is a unit that uses the service of ASR server separately from TCP socket and is distinguished by unique ID. ASR server supports using multiple channels with one TCP socket. The maximum number of available channels is limited by ASR server.

The ASR server saves the result data on its server before completing the requested voice recognition task and sending the result to the client. This data is used for continuous tuning of the voice recognition engine.

The ASR server runs through the console window, and the executed program does not terminate normally and continues to run. At this time, you can check the current channel status, I / O, and voice recognition results through the executed console window. (See Fig. 3)

3. DSR API

The DSR API performs the following functions.

(1) create a voice recording

When the recording request signal is received, WIPI's recording API is used to record the sound input from the microphone. The recorded sound source is encoded and stored in EVRC using the DSP of the terminal.

(2) Voice start point endpoint detection

The user presses the button once at the beginning of the talk and continues recording even after the button is released. Recording stops automatically when the user does not talk for a certain time. The captured sound source detects the start point and the end point based on the Zero Crossing Rate and dB.

(3) Voice Level Display

When the voice recording is in progress, a small icon indicating the volume of the voice recorded at a specific position on the terminal screen is displayed.

Given.

I. interface

An example of an on-demand voice navigation service provided based on such a platform is as follows.

When voice recognition is enabled, a lip icon is displayed. When the user presses the voice input button ("call" button), the volume of the voice being recorded is expressed in the form of an icon at the bottom of the screen. You can then speak your voice command for the desired wireless site / content. For example, if you say "Alo Homora" and the voice recording is completed and the voice is transferred to the DSR, the voice recognition icon is placed at the bottom of the screen so that the user can check the progress of the voice recognition. Voice recording is started with confirmation sound when it is pressed from off state to on state because there is a separate recognition hotkey. At this time, the progressive bar indicating the remaining time of voice recording is displayed for about 2 or 3 seconds and the end point detection is also performed.

If the recognition is successful, the relevant site / content title, for example, "Peep Mobile GoStop" appears on the screen, allowing the user to feed back the recognition result, and if the recognition result is incorrect, the user can try again before connecting. If you leave it alone, you will be directed to the site / content.

The greatest advantage of the present invention is that the user visits the wireless site / content at a time through the voice command, and uses the voice command in an interesting order and spell form to increase the voice recognition performance and to activate it as a commercial service. This will increase magic, magic, and related entertainment, making it fun to enter a lot of users by spelling it out, and also increasing the speech recognition rate by the length of spells and the uniqueness / differentiation of syllables. Created and widespread orders can brand the wireless site / content and be used for major advertising marketing. For example, while launching the "Green Pepper Mobile GoStop" service, call the ad, "Press the voice recognition call button on your mobile phone and say 'Alo Homora" and the new service will be in front of you. " Would be

Claims (3)

Wireless Internet Voice Navigation Technology Using Multi-modal Voice Recognition Platform Wireless internet voice navigation service through incantation type voice command Divisional voice command registration service for wireless Internet site / contents

Images