TWI783084B - Method and system of weight-based usage model for dynamic speech recognition channel selection - Google Patents

Abstract

The disclosure provides a method and system of weight-based usage model for dynamic speech recognition channel selection. The system includes a speech recognition channel setting module, a speech recognition channel database, a speech recognition channel module, a speech recognition channel weighting calculating module, a speech recognition channel capacity calculating module, a speech recognition channel decision module, and a speech recognition channel usage calculating module.

Description

Method and system for applying weight usage model to dynamically select speech recognition channel

The present invention relates to a speech recognition method and system, and in particular to a method and system for dynamically selecting a speech recognition channel using a weight usage model.

In the era of artificial intelligence, speech recognition is an indispensable artificial intelligence recognition technology in life. Speech recognition has its own unique domain, and it cannot provide an all-in-one solution. In addition, its training cost is too high, making it unaffordable for corporate or personal users.

Recent related patents: Speech recognition system using multi-grammar network (Republic of China patent number: I394926, application date: 2000/06/21), this article proposes a multilingual speech recognition method, the steps are: receive sound frame; plural Speech models of different language families; entering the speech model according to the sound frame into a plurality of speech state scores corresponding to a plurality of speech states; selecting a plurality of the highest state scores from these speech state scores; and generating according to the state scores correction value. The multilingual speech recognition method eliminates the phenomenon of multilingual deviation in a dynamic and individual (Frame by Frame) manner.

Recent related patents: multilingual speech recognition device and method (Republic of China patent I579829, 2017/04/21), this patent proposes a multilingual speech recognition method, which may include the following steps: receiving sound frame; multiple different language families The voice model of the voice; according to the sound frame entering the voice model, a plurality of voice state scores corresponding to the multiple voice states are generated; selecting a plurality of the highest state scores from the voice state scores; and generating correction values according to the state scores . This multilingual speech recognition method eliminates the phenomenon of multilingual deviation in a dynamic (Dynamic) and individual (Frame by Frame) manner.

The above patents that are similar to the present invention still have the following disadvantages: (1) provide dynamic switching without considering the needs of the user's voice recognition; and (2) do not consider dynamically updating the usage of the voice recognition channel to improve the voice selected by the user for the next calculation Identify the accuracy of the channel.

The recent related U.S. patent number US9569174 B2, the invention name is " Methods and systems for managing speech recognition in a multi-speech system environment " (application date: 2017/02/17), this patent proposes to manage voice in a multi-speech system environment Management method and system, select the voice recognition system through the user's voice, gesture, and gaze. This patent proposes to record the user's actions, voice, and gesture modules, and the processor judges the selection based on the records of the user's actions, voice, and gesture modules. voice recognition system.

The above patents similar to the present invention still have the following disadvantages: (1) The voice recognition channel weight and the user's voice recognition channel usage are not considered; and (2) The voice recognition channel capacity is not considered, and the voice recognition channel cannot be dynamically configured according to the channel capacity resource.

It can be seen that the above-mentioned customary method still has many deficiencies, which is not a good design and needs to be improved urgently. In view of the shortcomings derived from the above-mentioned conventional methods, the inventor of this case is eager to improve and innovate. After years of painstaking research, he finally successfully developed and completed the method and system for dynamically selecting speech recognition channels using a weight usage model. .

The present invention provides an all-in-one voice recognition channel selection solution that can achieve the above-mentioned purpose of the invention, which refers to a method and system for dynamically selecting voice recognition channels using a weight usage model. The aforementioned system includes: (1) voice recognition channel setting module, used for channel and channel weight setting and access; (2) voice recognition channel module, used to receive user voice data and channel strategy data, and perform channel Switching and returning the recognition result data; (3) voice recognition channel weight calculation module, used to calculate the weight value of the voice recognition channel; (4) voice recognition channel capacity calculation module, used to dynamically calculate the voice recognition channel capacity; ( 5) Speech recognition channel decision-making module, which is used to decide and sort the results of the voice recognition channel selected by the user; (6) Speech recognition channel usage calculation module, which is used to dynamically update the channel weight according to the user's channel usage. The group can meet the user's service considerations and improve the dynamic resource allocation and allocation accuracy of the speech recognition channel.

In order to make the above-mentioned features and advantages of the present invention more comprehensible, the following specific embodiments are described in detail together with the accompanying drawings.

In the era of artificial intelligence, speech recognition is an indispensable artificial intelligence recognition technology in life. Speech recognition has its own unique domain and cannot provide an all-in-one solution. In addition, its high training cost makes enterprises Or if the user for personal use cannot afford the cost, this method provides dynamic switching of the voice recognition channel based on the user's voice recognition channel setting, voice recognition channel capacity, and user voice recognition channel usage.

Please refer to Fig. 1, which is a structure diagram of a method and a system for dynamically selecting speech recognition channels based on a weight usage model of the present invention. The method of the present invention can effectively improve resource allocation of speech recognition channels. The system includes: speech recognition Channel setting module 1 is for setting voice recognition channels and channel weights and accessing channel setting data; voice recognition channel database 2 is for storing voice recognition channel data; voice recognition channel module 3 is for receiving users Voice data and channel strategy data, and switch channels and return the identification result data; the voice recognition channel weight calculation module 4 is to calculate the channel weight and transmit and calculate the channel weight value; the voice recognition channel capacity calculation module 5 is the It is used to calculate channel capacity and transfer channel data; the speech recognition channel decision module 6 is to calculate the channel decision result and perform channel decision sorting; the speech recognition channel usage calculation module 7 is to calculate the user's channel usage And update the channel weight value.

Please refer to FIG. 2 , which is a component diagram for establishing a weight usage model for the method and system for dynamically selecting speech recognition channels by applying a weight usage model according to the present invention.

表1First of all, it is necessary to set the voice recognition channel strategy (Method: _{M j} , where j=1,2,...,N represents the voice recognition channel code), Table 1 shows that the user chooses this voice recognition channel strategy, 0 means not to choose this strategy, 1 means to choose this strategy, and the channel setting element 11 will set the voice The identification channel setting data is written into the speech recognition channel database 2, and the content of the data writing is as follows in Table 1:

Table 1

表2Then, through the channel weight setting element 12 of the voice recognition channel setting module 1, set the voice recognition channel strategy weight (Weight: _{WM j} , where j=1,2,...,N voice recognition channel code), the channel weight setting element 12 will set the user's voice recognition channel weight (WM _{1, UM} means that the Mth User sets the weight of voice recognition channel strategy 1 Value) setting data is written into the speech recognition channel database 2, and the content of the data writing is as follows in Table 2:

Table 2

（1）

（2）When the voice data is sent to the system, the voice data is received by the channel strategy component 31 of the voice recognition channel module 3, and the user information for transmitting the voice data is taken out, and the channel setting access component 13 of the voice recognition channel setting module 1 is used. Go to the speech recognition channel database to take out the speech recognition channel setting data and the speech recognition channel weight setting data, and send the fetched data to the weight data transmission component 41 of the speech recognition channel weight calculation module 4, and the weight calculation component 42 is responsible for calculating the user's voice Recognition channel weight, voice recognition channel weight calculation formula is as follows:

(1)

(2)

Equation (1) is the calculation formula for user 1 using speech recognition channel strategy 1, W' _{M1, U1} represents the weight calculation result of user 1 speech recognition channel strategy 1, and equation (2) is the speech recognition channel strategy for user 1 The calculation formula of N, W' _{MN, U1} represents the weight calculation result of user 1's speech recognition channel strategy N.

（3） …

（4）The weight data transmission component 41 transmits the weight calculation results of each voice recognition channel to the channel data transmission component 51 of the voice recognition capacity calculation module 5, and the capacity calculation component 52 is responsible for real-time calculation of the voice recognition channel capacity, and the calculation formula is as follows:

(3)…

(4)

Equation (3) is the channel capacity of voice recognition channel 1, where C _CH1 is the total number of channels available for voice recognition channel 1, C _{CH1, R} is the remaining number of channels available for voice recognition channel 1, and formula (4) is voice recognition channel N The channel capacity of , where C _CHN is the total number of channels available for speech recognition channel N, and C _CHN,R is the remaining number of available channels for speech recognition channel N.

表3Channel data transfer component 51 to channel usage update component 71 of voice recognition channel usage calculation module 7 to obtain real-time channel usage data, and channel usage update component 71 to voice recognition channel database 2 to obtain channel usage data, as shown in the following table 3:

table 3

（5） …

（6）The channel usage update component 71 transmits the data in Table 3 to the channel usage weight calculation component 72 to calculate the channel usage weight, and the calculation formula is as follows:

(5)…

(6)

Formula (5) is the usage weight of user 1 using speech recognition channel strategy 1, where P _{M1, U1} is the probability of user 1 using speech recognition channel strategy 1, and formula (6) is user 1 using speech recognition channel strategy The usage weight of N, where P _MN,U1 is the probability that user 1 uses speech recognition channel strategy N.

（7）

（8）The channel data transmission component 51 transmits the voice recognition channel weight calculation result, the real-time voice recognition channel capacity result and the channel usage weight calculation result to the decision sorting component 61 of the voice recognition channel decision module 6, and the decision sorting component 61 first transmits the data To the decision calculation element 62 to calculate the decision result, the calculation formula is as follows:

(7)

(8)

為使用者1的語音辨識通道1的決策結果，

為使用者1的語音辨識通道N的決策結果，決策排序元件61將排序

、…、

的結果，輸出結果以陣列方式排序{

, … ,

}，數值最高的結果排序於第一位，以D’ = {

, … ,

}表示決策結果數值陣列，其中CM表示決策結果陣列中元素個數，排序最後一個以1表示，排序最高以CM表示，即D’={CM,CM-1,…,1}。決策的結果數值若為0則不排列於結果陣列中；決策排序元件61將結果陣列傳送至語音辨識通道模組3的通道切換元件32，通道切換元件依據結果陣列進行切換，將辨識結果資料輸出。In formula (7)

is the decision result of user 1's speech recognition channel 1,

For the decision result of user 1's voice recognition channel N, the decision sorting component 61 will sort

,...,

The results, the output results are sorted in array {

, … ,

}, the result with the highest value is sorted first, with D' = {

, … ,

} represents the numerical array of decision results, where CM represents the number of elements in the decision result array, the last one is represented by 1, and the highest rank is represented by CM, that is, D'={CM,CM-1,…,1}. If the result value of the decision is 0, it will not be arranged in the result array; the decision sorting component 61 transmits the result array to the channel switching component 32 of the voice recognition channel module 3, and the channel switching component switches according to the result array, and outputs the recognition result data .

, … ,

}傳送至語音辨識通道使用量運算模組7的通道使用量更新元件71，通道使用量更新元件71至語音辨識通道資料庫2取得通道使用量記錄如表3，再將記錄傳送至通道使用量權重計算元件72進行使用量更新，語音辨識通道使用量更新的公式如下：

（9） …

（10）The decision sorting component 61 will make the decision result numerical array {

, … ,

} sent to the channel usage update component 71 of the voice recognition channel usage calculation module 7, the channel usage update component 71 is sent to the voice recognition channel database 2 to obtain the channel usage record as shown in Table 3, and then the record is sent to the channel usage The weight calculation component 72 updates the usage amount, and the formula for updating the usage amount of the speech recognition channel is as follows:

(9) …

(10)

P’_MN,U1 為使用者1的語音辨識通道N之使用量更新數值，表3為經過使用量計算元件72更新數據如表4，通道使用量更新元件71將表五結果儲存至語音辨識通道資料庫2。

表4Equation (9) is the update value of the usage of P' _{M1, U1} user 1's voice recognition channel 1, Equation (10)

P' _{MN, U1} is the update value of the usage of the voice recognition channel N of user 1, Table 3 shows the update data of the usage calculation component 72 as shown in Table 4, and the channel usage update component 71 stores the results of Table 5 to the voice recognition channel database2.

Table 4

Please refer to FIG. 3 , which is a system flowchart of a method and system for dynamically selecting speech recognition channels using a weight usage model according to the present invention. Firstly, in step 101, the channel policy component 31 notifies the channel setting access component 13 to the voice recognition channel database 2 to retrieve the voice recognition channel setting data. Step 102 judges whether there is voice recognition channel setting data. If not, step 103 must be executed first, the voice recognition channel data is set by the channel setting component 11, then step 104 channel weight setting component 12 sets the voice recognition channel weight data, and step 105 writes the voice recognition channel data and the voice recognition channel weight data If so, the channel strategy component 31 transmits the voice recognition channel data to the weight data transmission component 41, and then transmits it to the weight calculation component 42 to perform step 106 to calculate the voice recognition channel weight. The calculation result needs to be transmitted to the channel data transmission component 51 through the weight data transmission component 41 , and then transmitted to the capacity calculation component 52 to perform step 107 to calculate the voice recognition channel capacity. The channel data transmission component 51 transmits the voice recognition channel usage data to the channel usage updating component 71 , and then to the channel usage weight calculation component 72 . Execute step 108 to calculate the voice recognition channel usage, and finally the channel data transmission component 51 will calculate the voice recognition channel weight calculation results (calculation results of formula (1) and formula (2)), voice recognition channel capacity calculation results (formula (3) , the calculation result of formula (4), the calculation result of speech recognition channel usage (the calculation result of formula (5), formula (6)) is transmitted to the decision sorting component 61 through the channel data transmission component 51, and then sent to 62 for decision calculation The component executes step 109 to calculate the speech recognition channel decision. The decision calculation results (the calculation results of formula (7) and formula (8)) will be sorted by the decision sorting component 61 and the sorting results will be sent to the channel usage updating component 71 to perform step 110 to update the speech recognition channel usage (Formula (9), calculation results of formula (10)). The decision sorting component 61 transmits the sorting result to the channel switching component 32 to execute step 111 to switch the speech recognition channel, and finally outputs the speech recognition result data.

表5Taking 3 users and 3 voice recognition channel strategies as an example, the total number of channels for each strategy of the 3 voice recognition channel strategies is: strategy 1=4, strategy 2=4, strategy 3=2 channels; To set the voice recognition channel data, the user sets the voice recognition channel data through the channel setting component 11, as shown in Table 5 below, and then writes the set data into the voice recognition channel database 2:

table 5

表6Next, the user sets the speech recognition channel strategy weight through the channel weight setting component 12, as shown in Table 6 below, and then writes the set data into the speech recognition channel database 2:

Table 6

When user 1 sends voice data to this system, the channel policy element 31 of voice recognition channel module 3 receives the voice data, and takes out user 1's voice recognition channel setting data and voice recognition channel weight setting data, and takes out The setting data is sent to the weight data transmission component 41 of the voice recognition channel weight calculation module 4 . The weight calculation component 42 is responsible for calculating the weight of the speech recognition channel of user 1, and the calculation result is as follows:

= 0.25

= 0.75

= 1The weight data transmission component 41 transmits the calculation result of the voice recognition channel weight of the user 1 to the channel data transmission component 51 of the voice recognition capacity calculation module 5 . The capacity calculating component 52 is responsible for calculating the capacity of the speech recognition channel in real time. This example assumes that strategy 1 has a total of 4 channels, 1 channel is not used, strategy 2 has a total of 4 channels, unused 3 channels, and strategy 3 has a total of 2 channels. When no channel is used, the formula for calculating the channel capacity of the real-time voice recognition channel is as follows:

= 0.25

= 0.75

= 1

表7Channel data transfer component 51 to channel usage update component 71 of voice recognition channel usage calculation module 7 to obtain real-time channel usage data, and channel usage update component 71 to voice recognition channel database 2 to obtain channel usage data, as shown in the following table 7:

Table 7

The channel usage update component 71 transmits the data in Table 3 to the channel usage weight calculation component 72 to calculate the channel usage weight, and the calculation formula is as follows:

= 0.0975

.39The channel data transmission component 51 transmits the calculation result of the voice recognition channel weight of user 1, the real-time operation voice recognition channel capacity result and the channel usage weight result to the decision sorting component 61 of the voice recognition channel decision module 6, and the decision sorting component 61 first sends the data Send to the decision calculation element 62 to calculate the decision result, the calculation result is as follows:

= 0.0975

.39

、

、

的結果，輸出結果以陣列方式排序{

,

}={0.39,0.0975}，決策的結果

數值為0則不列於結果陣列中；決策結果數值陣列D’ = {

,

}={2,1}，決策排序元件61將排序結果傳送至語音辨識通道模組3的通道切換元件32，通道切換元件依據排序結果進行切換，將辨識結果資料輸出。Decision Sequencing Element 61 will sort

,

,

The results, the output results are sorted in array {

,

}={0.39,0.0975}, the result of the decision

If the value is 0, it will not be listed in the result array; the decision result value array D' = {

,

}={2,1}, the decision sorting component 61 transmits the sorting result to the channel switching component 32 of the voice recognition channel module 3, the channel switching component switches according to the sorting result, and outputs the recognition result data.

,

}={2,1}，其中共有兩個非0的方案，故CM=2，更新的公式如下：

=

3.33

= 2

=

1.67The decision sorting component 61 transmits the numerical array of decision results to the channel usage updating component 71 of the speech recognition channel usage calculation module 7 . The channel usage updating component 71 goes to the speech recognition channel database to obtain the channel usage record as shown in Table 7, and then sends the record to the channel usage weight calculation component 72 to update the usage, and the numerical result according to the decision result is {

,

}={2,1}, there are two non-zero solutions, so CM=2, the updated formula is as follows:

=

3.33

= 2

=

1.67

表8Write the above-mentioned channel usage update data into the speech recognition channel database 2 through the channel usage update component 71, and the updated data is shown in Table 8:

Table 8

Features and functions

A weight usage model provided by the present invention is applied to the method and system for dynamically selecting speech recognition channels. When compared with other conventional technologies, it has the following advantages:

1. The method of the present invention is based on the user channel setting data, voice recognition channel capacity and channel usage weight to determine the voice recognition channel, improve the accuracy of voice recognition channel switching, and effectively improve voice recognition under the consideration of user channel service satisfaction Channel resource configuration.

2. The voice recognition channel setting module of the present invention is used to set the voice recognition channel strategy selected by the user. The user can set multiple voice recognition channels according to the needs, providing users with multiple options.

3. The voice recognition channel module of the present invention is used to dynamically switch channels according to the decision result of the voice recognition channel.

4. The speech recognition channel weight calculation module of the present invention is used to calculate the weight of the speech recognition channel according to the user's channel weight setting value.

5. The voice recognition channel capacity calculation module of the present invention is used to dynamically calculate the voice recognition channel capacity, increase the voice recognition channel capacity, and improve the resource allocation of the voice recognition channel.

6. The voice recognition channel decision-making module of the present invention performs voice recognition channel decision-making calculation and sorting according to the voice recognition channel weight calculation, voice recognition channel capacity calculation module and voice recognition channel usage calculation module, considering user needs and Under the channel capacity, the dynamic resource allocation of the speech recognition channel is improved.

7. The speech recognition channel usage calculation module of the present invention is used to calculate the weight of the user channel usage, provide dynamic update of the user channel usage weight, and improve the accuracy of the user's voice channel recognition next time.

8. The present invention relates to a method and system for applying a weight usage model to dynamically select a voice recognition channel, especially the application of a voice recognition channel weight calculation module, a voice recognition channel capacity calculation module, and a voice recognition channel usage calculation module And voice recognition channel decision-making module, using this method to provide dynamic voice recognition channel selection according to user voice recognition channel settings, voice recognition channel capacity and voice recognition channel usage weight, improve the accuracy of switching voice recognition channels and effectively improve voice recognition Channel resource allocation.

The effects obtained by the present invention are as follows: (1) Through the voice recognition channel weight calculation module, the voice recognition channel capacity calculation module and the voice recognition channel usage calculation module, according to the user's voice recognition channel weight and voice recognition channel capacity dynamic Calculate the weight to dynamically configure the speech recognition channel in consideration of the service satisfaction of the user channel; (2) The speech recognition channel decision module switches the user's speech recognition channel according to the decision calculation result, effectively improving the resource allocation of the speech recognition channel (3 ) through the voice recognition channel usage calculation module to dynamically update the channel usage weight to improve the accuracy of switching voice recognition channels.

Although the present invention has been disclosed above with the embodiments, it is not intended to limit the present invention. Anyone with ordinary knowledge in the technical field may make some changes and modifications without departing from the spirit and scope of the present invention. The scope of protection of the present invention should be defined by the scope of the appended patent application.

1: Voice recognition channel setting module 11: Channel setting component 12: Channel weight setting component 13: Channel setting access component 2: Voice recognition channel database 3: Voice recognition channel module 31: Channel strategy component 32: Channel switching component 4: voice recognition channel weight calculation module 41: weight data transmission component 42: weight calculation component 5: voice recognition channel capacity calculation module 51: channel data transmission component 52: capacity calculation component 6: voice recognition channel decision module 61: Decision sorting component 62: Decision calculation component 7: Voice recognition channel usage calculation module 71: Channel usage updating component 72: Channel usage weight calculation components 101~111: A method for dynamically selecting a voice recognition channel using a weight usage model Flow chart of the system

FIG. 1 is a system architecture diagram of a method and system for dynamically selecting speech recognition channels using a weight usage model. FIG. 2 is a component diagram of a method and system for dynamically selecting speech recognition channels using a weight usage model. FIG. 3 is a flow chart of a method and system for dynamically selecting speech recognition channels using a weight usage model.

1: Speech recognition channel setting module

2: Speech recognition channel database

3: Speech recognition channel module

4: Speech recognition channel weight calculation module

5: Speech recognition channel capacity calculation module

6:Speech recognition channel decision module

7: Speech recognition channel usage calculation module

Claims (8)

A weight usage model is applied to a system for dynamically selecting speech recognition channels, including: a speech recognition channel setting module, which is used to set a user's selection strategy for multiple speech recognition channels and the strategy weight of each speech recognition channel, wherein If the ath user chooses the bth speech recognition channel strategy, set M _{b, Ua} to be 1, otherwise set _{Mb, Ua} to be 0, and the ath user has a strategy for the bth speech recognition channel The set strategy weight is represented as W _{Mb, Ua} ; a voice recognition channel database, used to store the selection strategy of each voice recognition channel and the strategy weight of each voice recognition channel; a voice recognition channel module, used To receive the voice data of the user, and receive the selection strategy of each of the voice recognition channels of the user and the strategy weight of each of the voice recognition channels retrieved by the voice recognition channel setting module from the voice recognition channel database ; A speech recognition channel weight calculation module, used to calculate the channel weight of each speech recognition channel based on the selection strategy of each speech recognition channel and the strategy weight of each speech recognition channel, and transmit the channel weight of each speech recognition channel The channel weight, wherein the channel weight corresponding to the a-th user and the b-th voice recognition channel strategy is W' _{Mb, Ua} , and

, where N is the number of these voice recognition channels; a voice recognition channel capacity computing module, which receives the channel weights of each of the voice recognition channels, and calculates the channel capacity of each of the voice recognition channels accordingly, wherein the kth voice The channel capacity of the identification channel is characterized by

, where C _CHk is the total number of channels available for the k-th voice recognition channel, C _{CHk, R} is the remaining number of available channels for the k-th voice recognition channel; a voice recognition channel usage calculation module for Obtain the real-time usage amount of each voice recognition channel, and calculate the channel usage weight of each voice recognition channel accordingly, wherein the real-time usage amount corresponding to the a-th user and the k-th voice recognition channel Represented as P _Mk,Ua , and the channel usage weight corresponding to the ath user and the kth speech recognition channel strategy is represented as

; A voice recognition channel decision module, used to calculate the decision result of each voice recognition channel based on the channel capacity, the channel weight and the channel usage weight of each voice recognition channel, and sort the voice recognition channels accordingly To generate a result array, wherein the voice recognition channel decision module provides the result array to the voice recognition channel usage calculation module and the voice recognition channel module, wherein the ath user is for the kth voice The decision result of the identification channel is characterized by D _Ua,CHk = W' _Mk,Ua ×

, and the result array corresponding to the a-th user is represented as { D' _{U1 , CH1} ,..., D' _{U1 , CHN} }, wherein each element in the result array is not 0, and the number is CM; the speech recognition channel usage calculation module calculates the usage update value of each of the speech recognition channels, wherein the usage update value of the kth speech recognition channel of the a user is represented as

; Wherein, the voice recognition channel module switches among the voice recognition channels according to the result array, and returns the voice recognition result. The system as described in item 1 of the scope of the patent application, wherein the speech recognition channel setting module includes: a channel setting element, which is used to provide the user with settings for each of the speech recognition channels The selection strategy of the channel, and write it into the voice recognition channel database; a channel weight setting element, used for the user to set the strategy weight of each voice recognition channel, and write it into the voice recognition channel accordingly database; a channel setting access element, used for accessing and transmitting the selection strategy of each speech recognition channel and the strategy weight of each speech recognition channel. The system described in item 1 of the scope of patent application, wherein the voice recognition channel module includes: a channel strategy element for receiving the voice data of the user, and receiving the user from the voice recognition channel setting module The selection strategy of each of the speech recognition channels and the strategy weight of each of the speech recognition channels; a channel switching element, used to receive the result array of the speech recognition channel decision module, and according to these speech recognition channels switch between. The system described in item 1 of the scope of the patent application, wherein the speech recognition channel weight calculation module includes: a weight calculation element for calculating the channel weight of each of the speech recognition channels; a weight data transmission element for transmitting The channel weight of each speech recognition channel. The system described in item 1 of the scope of the patent application, wherein the voice recognition channel capacity calculation module includes: a channel data transmission component, responsible for receiving the channel weight of each voice recognition channel and forwarding it to the voice recognition channel The usage calculation module and the voice recognition a recognition channel decision module; and a capacity calculation element, which is used to dynamically calculate the channel capacity of each speech recognition channel. The system as described in item 1 of the scope of patent application, wherein the voice recognition channel decision module includes: a decision calculation element, which is used to calculate the channel capacity, the channel weight result and the channel usage weight of each voice recognition channel calculating the decision result of each of the speech recognition channels; and a decision sorting component for sorting the decision results of each of the speech recognition channels to generate the result array, and sending the result array to the speech recognition channel module and the Speech recognition channel usage calculation module. The system as described in item 1 of the scope of the patent application, wherein the voice recognition channel usage calculation module includes: a channel usage weight calculation element for calculating the channel usage weight of each voice recognition channel; and a channel The usage updating component is used for dynamically receiving the channel usage weight of each voice recognition channel from the channel usage weight calculation component, and updating the voice recognition channel database accordingly. A kind of weight consumption model is applied to the method for dynamically selecting the voice recognition channel, and its process steps are: A. set a user's selection strategy to a plurality of voice recognition channels by the voice recognition channel setting element; B. set the voice recognition channel weight by the element Set the strategy weight of each speech recognition channel, wherein if the ath user selects the bth speech recognition channel strategy, set _{Mb, Ua} to 1, otherwise set _{Mb, Ua} to 0, and the ath uses Or the strategy weight set for the b-th voice recognition channel strategy is represented as W _{Mb, Ua} : C. transmit the voice data of the user to the channel strategy element; D. call the channel setting and access by the channel strategy element The component uses the selection strategy and the strategy weight of each speech recognition channel; E. Set the access component from the channel to a speech recognition channel database to obtain the selection strategy and the strategy weight of each speech recognition channel; F. The selection strategy and the strategy weight of each of the speech recognition channels are transmitted to the weight data transmission element, and then the weight calculation element calculates the channel weight of each of the speech recognition channels, which corresponds to the ath user and the ath user The channel weight of the b speech recognition channel strategy is W' _{Mb, Ua} , and W' _{Mb , Ua} =

, wherein N is the quantity of these voice recognition channels; G. The channel weight of each of the voice recognition channels is transmitted to the channel data transmission element; H. The capacity calculation element of the voice recognition channel is called by the channel data transmission element for real-time dynamics Calculate the channel capacity of each speech recognition channel, wherein the channel capacity of the kth speech recognition channel is represented as

, where C _CHk is the total number of available channels of the k-th voice recognition channel, C _{CHk, R} is the remaining available channel number of the k-th voice recognition channel; 1. is updated by channel data transmission element call channel usage The component obtains the real-time usage amount of each voice recognition channel, and calculates the channel usage weight of each voice recognition channel, wherein the real-time usage amount corresponding to the a-th user and the k-th voice recognition channel represents is P _Mk,Ua , and the channel usage weight corresponding to the ath user and the kth speech recognition channel strategy is expressed as

J. The channel capacity, the channel weight and the channel usage weight of each of the speech recognition channels are transmitted to the decision-making sequence element by the channel data transmission element; The decision result of the recognition channel, wherein the decision result of the ath user for the kth speech recognition channel is represented as

, and the result array corresponding to the a-th user is represented as { D' _{U1 , CH1} ,..., D' _{U1 , CHN} }, wherein each element in the result array is not 0, and the number is CM; L. The decision-making sorting component transmits the decision result of each voice recognition channel to the channel usage update component for calculation, and updates the channel of each voice recognition channel according to the real-time usage of each voice recognition channel weight, and update to the speech recognition channel database accordingly, wherein the usage update value of the kth speech recognition channel of the ath user is represented as

M. The decision ordering component transmits the decision result of each voice recognition channel to the channel switching component; N. The channel switching component switches among the voice recognition channels and returns the voice recognition result.

Images