Human-machine interactive speech recognizing method and system for intelligent devices

Abstract

A human-machine interactive speech recognizing method and system for intelligent devices, pertaining to the technical field of speech recognition, and enhancing accuracy of speech recognition by jointly optimizing and training intent recognition and slot filling. The method comprises: subjecting user's speech question to a term-segmenting process to obtain original term sequence, and vectorizing original term sequence through embedding process; weighting hidden state vector hi and slot context vector ciS to obtain slot label model yiS; weighting hidden state vector hT and intent context vector cI to obtain intent prediction model yI; employing slot gate g to join slot context vector ciS and intent context vector cI, and generating transformed representation of slot label model yiS through slot gate g; and jointly optimizing intent prediction model yI and transformed slot label model yiS to construct target function, and performing intent recognition on user's speech question based on target function.

Description

HUMAN-MACHINE INTERACTIVE SPEECH RECOGNIZING METHOD AND SYSTEM FOR INTELLIGENT DEVICES BACKGROUND OF THE INVENTION Technical Field

[0001] The present invention relates to the technical field of speech recognition, and more particularly to a human-machine interactive speech recognizing method and system for an intelligent device. Description of Related Art

[0002] With the development of the internet technology, there come into being more and more intelligent devices that employ speeches for human-machine interaction. Currently available speech interactive systems include Siri, Xiaomi, Cortana, Avatar Framework, and Duer, etc. As compared with the traditional human-machine interaction based on manual input, speech human-machine interaction exhibits characteristics of conveniency, high efficiency, and broad range of application scenarios. During the process of speech recognition, intent recognition and slot filling techniques are keys to ensuring the accuracy of speech recognition results.

[0003] As regards intent recognition, it can be abstracted as a classification problem, and a classifier represented by means of CNN + knowledge is then employed to train an intent recognition model, in which is further introduced semantic representation of knowledge to enhance the generalization capability of the presentation layer in addition to word- embedding for speech questions of users, but it has been found in practical application that such a model is defective in terms of slot information filling deviation, whereby accuracy of the intent recognition model is adversely affected. As regards slot filling, its essence is to formalize a sentence sequence to a marked sequence, and there are many frequently used methods to mark sequences, such as the hidden Markov model or the conditional random field model, but these slot filling models cannot satisfy practical application requirements under specific application scenarios, due to ambiguities of slots existent under different semantic intents caused by the lack of contextual information. Seen as such, trainings of the two models are independently carried out in the state of the art, and there is no combined optimization of the intent recognition task and the slot filling task, so that the finally trained models are problematic in terms of low recognition accuracy in the aspect of speech recognition, and user experience is lowered. SUMMARY OF THE INVENTION

[0004] The objective of the present invention is to provide a human-machine interactive speech recognizing method and system for an intelligent device, to enhance accuracy of speech recognition by jointly optimizing and training intent recognition and slot filling.

[0005] To achieve the above objective, according to one aspect, the present invention provides a human-machine interactive speech recognizing method for an intelligent device, the method comprising:

[0006] subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process;

[0007] calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics>;

[0008] calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cl<annotation encoding="application / x-tex">c^{l}< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yl<annotation encoding="application / x-tex">y^{l}< / annotation>< / semantics>;

[0009] employing a slot gate g to join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>, and generating a transformed representation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate g; and

[0010] jointly optimizing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> to construct a target function, and performing intent recognition on the speech question of the user based on the target function.

[0011] Preferably, the step of subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process includes:

[0012] receiving the speech question of the user and transforming the speech question to a recognizable text, and employing a tokenizer to term-segment the recognizable text and obtain the original term sequence; and

[0013] subjecting the original term sequence to a word embedding process, and realizing a vector representation of each segmented term in the original term sequence.

[0014] Preferably, the step of calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> includes:

[0015] employing a bidirectional LSTM network to encode each term segmentation vector, and outputting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> corresponding to each term segmentation vector;

[0016] calculating the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>, to which each term segmentation vector corresponds, through formula <semantics>ciS=∑ai,jShj<annotation encoding="application / x-tex">c_i^S = \sum a_{i,j}^S h_j< / annotation>< / semantics>, wherein <semantics>ai,jS<annotation encoding="application / x-tex">a_{i,j}^S< / annotation>< / semantics> represents an attention weight of a slot, its calculation formula is <semantics>ai,jS=exp(ei,j)∑k=1Texp(ei,k),ei,j=σ(WheShj)<annotation encoding="application / x-tex">a_{i,j}^S = \frac{\exp(e_{i,j})}{\sum_{k=1}^T \exp(e_{i,k})}, e_{i,j} = \sigma(W_{he}^S hj)< / annotation>< / semantics>, where <semantics>σ<annotation encoding="application / x-tex">\sigma< / annotation>< / semantics> represents a slot activation function, and <semantics>WheS<annotation encoding="application / x-tex">W_{he}^{S}< / annotation>< / semantics> represents a slot weight matrix; and

[0017] constructing a slot label model <semantics>yiS=softmax(WheS(hi+ciS))<annotation encoding="application / x-tex">y_i^S = softmax (W_{he}^S(hi + c_i^S))< / annotation>< / semantics> based on the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>.

[0018] Further, the step of calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> includes:

[0019] employing a hidden unit in the bidirectional LSTM network to encode the vectorized original term sequence, and obtaining the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics>;

[0020] calculating the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the original term sequence through formula <semantics>cI=∑aiIhT<annotation encoding="application / x-tex">c^{I} = \sum a_{i}^{I} hT< / annotation>< / semantics>, wherein <semantics>aiI<annotation encoding="application / x-tex">a_{i}^{I}< / annotation>< / semantics> represents an attention weight of an intent, its calculation formula is <semantics>ajI=exp(ej)∑k=1Texp(ek)<annotation encoding="application / x-tex">a_j^I = \frac{\exp(e_j)}{\sum_{k=1}^T \exp(e_k)}< / annotation>< / semantics>, <semantics>ei=σ′(WhvIhT)<annotation encoding="application / x-tex">e_i = \sigma'(W_{hv}^I h T)< / annotation>< / semantics>, where <semantics>σ′<annotation encoding="application / x-tex">\sigma'< / annotation>< / semantics> represents an intent activation function, and <semantics>WhνI<annotation encoding="application / x-tex">W_{h\nu}^{I}< / annotation>< / semantics> represents an intent weight matrix; and

[0021] constructing an intent prediction model <semantics>yI=softmax(WhvI(hT+cI))<annotation encoding="application / x-tex">y^I = softmax(W_{hv}^I(hT + c^I))< / annotation>< / semantics> based on the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics>.

[0022] Preferably, the step of employing a slot gate g to join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics>, and generating a transformed representation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate g includes:

[0023] formally representing the slot gate g as <semantics>g=∑v⋅tanh<annotation encoding="application / x-tex">g = \sum v \cdot tanh< / annotation>< / semantics> <semantics>(cis+W⋅cI)<annotation encoding="application / x-tex">(c_i^s + W \cdot c^I)< / annotation>< / semantics>, wherein v represents a weight vector obtained by training, and W represents a weight matrix obtained by training; and

[0024] formally representing the transformation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate g as: [Image disponible dans le document PDF, Image available in the PDF document]

[0026] Optionally, the target function constructed by jointly optimizing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> is:

[0027] <semantics>p(yS,yI|X)=p(yI|X)∏p(yiS|X)<annotation encoding="application / x-tex">p(y^S, y^I|X) = p(y^I|X) \prod p(y_i^S|X)< / annotation>< / semantics>, wherein <semantics>p(yS,yI|X)<annotation encoding="application / x-tex">p(y^S, y^I|X)< / annotation>< / semantics> represents a conditional probability for outputting slot filling and intent prediction at a given original term sequence, where X is the vectorized original term sequence.

[0028] Preferably, the step of performing intent recognition on the speech question of the user based on the target function includes:

[0029] sequentially obtaining intent conditional probabilities, to which the various segmented terms in the original term sequence correspond, through the target function; and

[0030] screening therefrom a segmented term with the maximum probability value and recognizing the segmented term as the intent of the speech question of the user.

[0031] In comparison with prior-art technology, the human-machine interactive speech recognizing method for an intelligent device provided by the present invention achieves the following advantageous effects.

[0032] In the human-machine interactive speech recognizing method for an intelligent device provided by the present invention, the speech question of the user as obtained is firstly transformed to a recognizable text, a term segmenting process is carried out on the basis of the recognizable text to generate an original term sequence, which is then subjected to a word embedding process to realize vector representation, thereafter, a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> and an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> are respectively constructed on the basis of the vectorized original term sequence, wherein the step of constructing the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> is to calculate a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weight the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics>, while the step of constructing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> is to calculate a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the original term sequence, and weight the hidden state vector hT and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> to thereafter obtain the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^{I}< / annotation>< / semantics>; seen as such, in order to fuse the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> with the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics>, a decoder layer is additionally added to the existing encoder-decoder framework to construct the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^{I}< / annotation>< / semantics>, join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_{i}^{S}< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> by introducing a slot gate g, finally jointly optimize the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> to obtain a target function, employ the target function to sequentially obtain intent conditional probabilities, to which the various segmented terms in the original term sequence correspond, and screen therefrom a segmented term with the maximum probability value and recognize it as the intent of the speech question of the user, so as to ensure accuracy of speech recognition.

[0033] According to another aspect, the present invention provides a human-machine interactive speech recognizing system for an intelligent device, wherein the system is applied to the human-machine interactive speech recognizing method for an intelligent device as recited in the foregoing technical solution, and the system comprises:

[0034] a term segmentation processing unit, for subjecting a speech question of a user to a term- segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process;

[0035] a first calculating unit, for calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics>;

[0036] a second calculating unit, for calculating a hidden state vector hT and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics>;

[0037] a model transforming unit, for employing a slot gate g to join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics>, and generating a transformed representation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate <semantics>g<annotation encoding="application / x-tex">g< / annotation>< / semantics>; and

[0038] a joint optimization unit, for jointly optimizing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> to construct a target function, and performing intent recognition on the speech question of the user based on the target function.

[0039] Preferably, the term segmentation processing unit includes:

[0040] a term-segmenting module, for receiving the speech question of the user and transforming the speech question to a recognizable text, and employing a tokenizer to term-segment the recognizable text and obtain the original term sequence; and

[0041] an embedding processing module, for subjecting the original term sequence to a word embedding process, and realizing a vector representation of each segmented term in the original term sequence.

[0042] Preferably, the first calculating unit includes:

[0043] a hidden state calculating module, for employing a bidirectional LSTM network to encode each term segmentation vector, and outputting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> corresponding to each term segmentation vector;

[0044] a slot context calculating module, for calculating the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>, to which each term segmentation vector corresponds, through formula <semantics>ciS=∑ai,jShj<annotation encoding="application / x-tex">c_i^S = \sum a_{i,j}^S h_j< / annotation>< / semantics>, wherein <semantics>ai,jS<annotation encoding="application / x-tex">a_{i,j}^S< / annotation>< / semantics> represents an attention weight of a slot, its calculation formula is <semantics>ai,jS=exp(ei,j)∑k=1Texp(ei,k)<annotation encoding="application / x-tex">a_{i,j}^S = \frac{\exp(e_{i,j})}{\sum_{k=1}^T \exp(e_{i,k})}< / annotation>< / semantics>, <semantics>ei,i=σ(WheShj)<annotation encoding="application / x-tex">e_{i,i} = \sigma(W_{he}^S hj)< / annotation>< / semantics>, where <semantics>σ<annotation encoding="application / x-tex">\sigma< / annotation>< / semantics> represents a slot activation function, and <semantics>WheS<annotation encoding="application / x-tex">W_{he}^S< / annotation>< / semantics> represents a slot weight matrix; and

[0045] a slot label model module, for constructing a slot label model <semantics>yiS=<annotation encoding="application / x-tex">y_i^S =< / annotation>< / semantics> <semantics>softmax<annotation encoding="application / x-tex">softmax< / annotation>< / semantics> <semantics>(WheS(hi+ciS))<annotation encoding="application / x-tex">(W_{he}^{S}(hi+c_{i}^{S}))< / annotation>< / semantics> based on the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_{i}< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>.

[0046] As compared with prior-art technology, the advantageous effects achieved by the human- machine interactive speech recognizing system for an intelligent device provided by the present invention are identical with the advantageous effects achievable by the human- machine interactive speech recognizing method for an intelligent device provided by the foregoing technical solution, so these are not redundantly described in this context. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The drawings described here are meant to provide further understanding of the present invention, and constitute part of the present invention. The exemplary embodiments of the present invention and the descriptions thereof are meant to explain the present invention, rather than to restrict the present invention. In the drawings:

[0048] Fig. 1 is a flowchart schematically illustrating the human-machine interactive speech recognizing method for an intelligent device in Embodiment 1 of the present invention;

[0049] Fig. 2 is an exemplary view illustrating encoder-decoder fusing model in Embodiment 1 of the present invention;

[0050] Fig. 3 is an exemplary view illustrating the slot gate g in Fig. 2; and

[0051] Fig. 4 is a block diagram illustrating the structure of the human-machine interactive speech recognizing system for an intelligent device in Embodiment 2 of the present invention.

[0052] Reference numerals:

[0053] 1 – term segmentation processing unit

[0054] 3 – second calculating unit

[0055] 5 – joint optimization unit 2 – first calculating unit 4 – model transforming unit DETAILED DESCRIPTION OF THE INVENTION

[0056] To make more lucid and clear the objectives, features and advantages of the present invention, the technical solutions in the embodiments of the present invention are clearly and comprehensively described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the embodiments as described are merely partial, rather than the entire, embodiments of the present invention. All other embodiments obtainable by persons ordinarily skilled in the art on the basis of the embodiments in the present invention without spending creative effort shall all fall within the protection scope of the present invention.

[0057] Embodiment 1

[0058] Fig. 1 is a flowchart schematically illustrating the human-machine interactive speech recognizing method for an intelligent device in Embodiment 1 of the present invention. Referring to Fig. 1, the human-machine interactive speech recognizing method for an intelligent device provided by this embodiment comprises:

[0059] subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process; calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics>; calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics>; employing a slot gate g to join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>, and generating a transformed representation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate g; and jointly optimizing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^{I}< / annotation>< / semantics> and the transformed slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> to construct a target function, and performing intent recognition on the speech question of the user based on the target function.

[0060] In the human-machine interactive speech recognizing method for an intelligent device provided by this embodiment, the speech question of the user as obtained is firstly transformed to a recognizable text, a term segmenting process is carried out on the basis of the recognizable text to generate an original term sequence, which is then subjected to a word embedding process to realize vector representation, thereafter, a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> and an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> are respectively constructed on the basis of the vectorized original term sequence, wherein the step of constructing the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> is to calculate a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weight the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics>, while the step of constructing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> is to calculate a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the original term sequence, and weight the hidden state vector hT and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics>; as shown in Fig. 2, in order to fuse the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> with the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics>, a decoder layer is additionally added to the existing encoder-decoder framework to construct the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics>, join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> by introducing a slot gate <semantics>g<annotation encoding="application / x-tex">g< / annotation>< / semantics>, finally jointly optimize the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> to obtain a target function, employ the target function to sequentially obtain intent conditional probabilities, to which the various segmented terms in the original term sequence correspond, and subsequently screen therefrom a segmented term with the maximum probability value and recognize it as the intent of the speech question of the user, so as to ensure accuracy of speech recognition.

[0061] Specifically, the step of subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process in the foregoing embodiment includes:

[0062] receiving the speech question of the user and transforming the speech question to a recognizable text, and employing a tokenizer to term-segment the recognizable text and obtain the original term sequence; and subjecting the original term sequence to a word embedding process, and realizing a vector representation of each segmented term in the original term sequence.

[0063] As should be noted, the step of calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> in the foregoing embodiment includes:

[0064] employing a bidirectional LSTM network to encode each term segmentation vector, and outputting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> corresponding to each term segmentation vector; calculating the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>, to which each term segmentation vector corresponds, through formula <semantics>ciS=∑ai,jShj<annotation encoding="application / x-tex">c_i^S = \sum a_{i,j}^S h_j< / annotation>< / semantics>, wherein <semantics>ai,jS<annotation encoding="application / x-tex">a_{i,j}^S< / annotation>< / semantics> represents an attention weight of a slot, its calculation formula is <semantics>ai,jS=exp(ei,j)∑b=1Texp(ei,b)<annotation encoding="application / x-tex">a_{i,j}^S = \frac{\exp(e_{i,j})}{\sum_{b=1}^T \exp(e_{i,b})}< / annotation>< / semantics>, <semantics>ei,j=σ(WheShj)<annotation encoding="application / x-tex">e_{i,j} = \sigma(W_{he}^S hj)< / annotation>< / semantics>, where <semantics>σ<annotation encoding="application / x-tex">\sigma< / annotation>< / semantics> represents a slot activation function, and <semantics>WheS<annotation encoding="application / x-tex">W_{he}^{S}< / annotation>< / semantics> represents a slot weight matrix; and constructing a slot label model <semantics>yiS=softmax(WheS(hi+ciS))<annotation encoding="application / x-tex">y_i^S = softmax (W_{he}^S(hi + c_i^S))< / annotation>< / semantics> based on the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>.

[0065] During specific implementation, after plural term segmentation vectors have been input to the bidirectional LSTM network, hidden state vectors <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> can be correspondingly output on a one-by-one basis, as regards formula <semantics>ciS=∑ai,jShj<annotation encoding="application / x-tex">c_i^S = \sum a_{i,j}^S h_j< / annotation>< / semantics> of the slot context vector, where <semantics>ai,iS<annotation encoding="application / x-tex">a_{i,i}^{S}< / annotation>< / semantics> represents the attention weight of the slot, i represents the ith term segmentation vector, and j represents the <semantics>jth<annotation encoding="application / x-tex">j^{th}< / annotation>< / semantics> element in the <semantics>ith<annotation encoding="application / x-tex">i^{th}< / annotation>< / semantics> term segmentation vector. Specifically, the calculation formula of the attention weight of the slot is <semantics>ai,jS=exp(ei,j)∑ν=1Texp(ei,k)<annotation encoding="application / x-tex">a_{i,j}^S = \frac{\exp{(e_{i,j})}}{\sum_{\nu=1}^T \exp{(e_{i,k})}}< / annotation>< / semantics>, <semantics>ei,j=exp(ei,j)∑ν=1Texp(ei,k)<annotation encoding="application / x-tex">e_{i,j} = \frac{\exp{(e_{i,j})}}{\sum_{\nu=1}^T \exp{(e_{i,k})}}< / annotation>< / semantics> <semantics>σ(WheShj)<annotation encoding="application / x-tex">\sigma(W_{he}^S hj)< / annotation>< / semantics>, where T represents the total number of elements in the term segmentation vector, and K represents the <semantics>Kth<annotation encoding="application / x-tex">K^{th}< / annotation>< / semantics> element in T. In addition, as regards slot activation function <semantics>σ<annotation encoding="application / x-tex">\sigma< / annotation>< / semantics> and slot weight matrix <semantics>WheS<annotation encoding="application / x-tex">W_{he}^{S}< / annotation>< / semantics>, these can be derived on the basis of vector matrix training of the original term sequence, and the specific training processes are conventional technical means frequently employed in this technical field, so these are not redundantly described in this embodiment.

[0066] The step of calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> in the foregoing embodiment includes:

[0067] employing a hidden unit in the bidirectional LSTM network to encode the vectorized original term sequence, and obtaining the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics>; calculating the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the original term sequence through formula <semantics>cI=∑aiIhT<annotation encoding="application / x-tex">c^I = \sum a_i^I hT< / annotation>< / semantics>, wherein <semantics>ajI<annotation encoding="application / x-tex">a_j^I< / annotation>< / semantics> represents an attention weight of an intent, its calculation formula is <semantics>ajI=exp(ej)∑k=1Texp(ek)<annotation encoding="application / x-tex">a_j^I = \frac{\exp{(e_j)}}{\sum_{k=1}^T \exp{(e_k)}}< / annotation>< / semantics>, <semantics>ei=σ′(WhvIhT)<annotation encoding="application / x-tex">e_i = \sigma'(W_{hv}^I h T)< / annotation>< / semantics>, where <semantics>σ′<annotation encoding="application / x-tex">\sigma'< / annotation>< / semantics> represents an intent activation function, and <semantics>WhvI<annotation encoding="application / x-tex">W_{hv}^I< / annotation>< / semantics> represents an intent weight matrix; and constructing an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^{I}< / annotation>< / semantics> = <semantics>softmax(WhvI(hT+cI))<annotation encoding="application / x-tex">softmax(W_{hv}^{I}(hT+c^{I}))< / annotation>< / semantics> based on the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>.

[0068] During the process of specific implementation, the method of training the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> is the same as the method of training the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics>, and the difference rests in the fact that the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> can be obtained merely by means of a hidden unit in the bidirectional LSTM network, after one-dimensional transformation of the vector matrix, formula <semantics>cI=∑aiIhT<annotation encoding="application / x-tex">c^I = \sum a_i^I hT< / annotation>< / semantics> is subsequently invoked to calculate the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the original term sequence, where <semantics>aiI<annotation encoding="application / x-tex">a_i^I< / annotation>< / semantics> represents an attention weight of an intent, its calculation formula is <semantics>ajI=exp(ej)∑k=1Texp(ek)<annotation encoding="application / x-tex">a_j^I = \frac{\exp(e_j)}{\sum_{k=1}^T \exp(e_k)}< / annotation>< / semantics>, <semantics>ei=σ′(WhvIhT)<annotation encoding="application / x-tex">e_i = \sigma'(W_{hv}^I h T)< / annotation>< / semantics>, wherein <semantics>σ′<annotation encoding="application / x-tex">\sigma'< / annotation>< / semantics> represents an intent activation function, and <semantics>WhνI<annotation encoding="application / x-tex">W_{h\nu}^{I}< / annotation>< / semantics> represents an intent weight matrix; as regards the intent activation function <semantics>σ′<annotation encoding="application / x-tex">\sigma'< / annotation>< / semantics> and the intent weight matrix <semantics>WhνI<annotation encoding="application / x-tex">W_{h\nu}^{I}< / annotation>< / semantics>, these can be derived on the basis of processed one-dimensional vector training, the specific training processes are conventional technical means frequently employed in this technical field, so these are not redundantly described in this embodiment.

[0069] Moreover, the step of employing a slot gate g to join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics>, and generating a transformed representation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate g in the foregoing embodiment includes:

[0070] formally representing the slot gate g as <semantics>g=∑v⋅tanh<annotation encoding="application / x-tex">g = \sum v \cdot tanh< / annotation>< / semantics> (<semantics>cis+W⋅cI<annotation encoding="application / x-tex">c_i^s + W \cdot c^I< / annotation>< / semantics>), wherein v represents a weight vector obtained by training, and W represents a weight matrix obtained by training; and formally representing the transformation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate g as <semantics>yiS=softmax(WheS(hi+ciS⋅g))<annotation encoding="application / x-tex">y_i^S = softmax(W_{he}^S(hi + c_i^S \cdot g))< / annotation>< / semantics>. Fig. 3 shows a structure model of the slot gate g.

[0071] Preferably, the target function constructed by jointly optimizing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> in the foregoing embodiment is:

[0072] <semantics>p(yS,yI|X)=p(yI|X)∏p(yiS|X)<annotation encoding="application / x-tex">p(y^S, y^I|X) = p(y^I|X) \prod p(y_i^S|X)< / annotation>< / semantics>, wherein <semantics>p(yS,yI|X)<annotation encoding="application / x-tex">p(y^S, y^I|X)< / annotation>< / semantics> represents a conditional probability for outputting slot filling and intent prediction at a given original term sequence, where X represents the vectorized original term sequence. After expansion, <semantics>p(yS,yI|X)=p(yI|X)∏iTp(yiS|X)=p(yI|X1,…XT)∏iTp(yiS|X1,…XT)<annotation encoding="application / x-tex">p(y^S, y^I|X) = p(y^I|X) \prod_i^T p(y_i^S|X) = p(y^I|X_1, \dots X_T) \prod_i^T p(y_i^S|X_1, \dots X_T)< / annotation>< / semantics>, where <semantics>Xi<annotation encoding="application / x-tex">X_i< / annotation>< / semantics> represents the <semantics>ith<annotation encoding="application / x-tex">i^{th}< / annotation>< / semantics> term segmentation vector, and <semantics>T<annotation encoding="application / x-tex">T< / annotation>< / semantics> represents the total number of term segmentation vectors. Through calculation of the target function can be obtained intent probability values of the various term segmentation vectors, and a segmented term with the maximum probability value is screened out of the various term segmentation vectors and recognized as the intent of the speech question of the user.

[0073] Embodiment 2

[0074] Referring to Fig. 1 and Fig. 4, this embodiment provides a human-machine interactive speech recognizing system for an intelligent device, the system comprising:

[0075] a term segmentation processing unit 1, for subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process;

[0076] a first calculating unit 2, for calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics>;

[0077] a second calculating unit 3, for calculating a hidden state vector hT and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics>;

[0078] a model transforming unit 4, for employing a slot gate g to join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics>, and generating a transformed representation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate g; and

[0079] a joint optimization unit 5, for jointly optimizing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> to construct a target function, and performing intent recognition on the speech question of the user based on the target function. Specifically, the term segmentation processing unit includes:

[0080] a term-segmenting module, for receiving the speech question of the user and transforming the speech question to a recognizable text, and employing a tokenizer to term-segment the recognizable text and obtain the original term sequence; and

[0081] an embedding processing module, for subjecting the original term sequence to a word embedding process, and realizing a vector representation of each segmented term in the original term sequence.

[0082] Specifically, the first calculating unit includes:

[0083] a hidden state calculating module, for employing a bidirectional LSTM network to encode each term segmentation vector, and outputting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> corresponding to each term segmentation vector;

[0084] a slot context calculating module, for calculating the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>, to which each term segmentation vector corresponds, through formula <semantics>ciS=∑ai,jShi<annotation encoding="application / x-tex">c_i^S = \sum a_{i,j}^S h_i< / annotation>< / semantics>, wherein <semantics>ai,jS<annotation encoding="application / x-tex">a_{i,j}^S< / annotation>< / semantics> represents an attention weight of a slot, its calculation formula is <semantics>ai,jS=exp(ei,j)∑k=1Texp(ei,k)<annotation encoding="application / x-tex">a_{i,j}^{S} = \frac{\exp(e_{i,j})}{\sum_{k=1}^{T} \exp(e_{i,k})}< / annotation>< / semantics>, <semantics>ei,j=σ(WheShj)<annotation encoding="application / x-tex">e_{i,j} = \sigma(W_{he}^S hj)< / annotation>< / semantics>, where <semantics>σ<annotation encoding="application / x-tex">\sigma< / annotation>< / semantics> represents a slot activation function, and <semantics>WheS<annotation encoding="application / x-tex">W_{he}^S< / annotation>< / semantics> represents a slot weight matrix; and

[0085] a slot label model module, for constructing a slot label model <semantics>yis=<annotation encoding="application / x-tex">y_i^s =< / annotation>< / semantics> <semantics>softmax<annotation encoding="application / x-tex">softmax< / annotation>< / semantics> <semantics>(WheS(hi+ciS))<annotation encoding="application / x-tex">(W_{he}^S(hi+c_i^S))< / annotation>< / semantics> based on the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>.

[0086] As compared with prior-art technology, the advantageous effects achieved by the human- machine interactive speech recognizing system for an intelligent device provided by this embodiment of the present invention are identical with the advantageous effects achievable by the human-machine interactive speech recognizing method for an intelligent device provided by the foregoing Embodiment 1, so these are not redundantly described in this context.

[0087] As understandable to persons ordinarily skilled in the art, realization of the entire or partial steps in the method of the present invention can be completed via a program that instructs relevant hardware, the program can be stored in a computer-readable storage medium, and subsumes the various steps of the method in the foregoing embodiment when it is executed, wherein the storage medium can be an ROM / RAM, a magnetic disk, an optical disk, or a memory card, etc.

[0088] What the above describes is merely directed to specific modes of execution of the present invention, but the protection scope of the present invention is not restricted thereby. Any change or replacement easily conceivable to persons skilled in the art within the technical range disclosed by the present invention shall be covered by the protection scope of the present invention. Accordingly, the protection scope of the present invention shall be based on the protection scope as claimed in the Claims.

Claims

<pat:ClaimStatement>What is claimed is:< / pat:ClaimStatement> <pat:Claims com:id="claims"> <pat:Claim com:id="CLM-00001"> <pat:ClaimNumber>1< / pat:ClaimNumber> <pat:ClaimText>1. A human-machine interactive speech recognizing method for an intelligent device, characterized in comprising: subjecting, by a term segmentation processing unit, a speech question of a user to a term- segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process; and calculating, by a first calculating unit, a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics>, wherein the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> is calculated by encoding each term segmentation vector by a bidirectional long short- term memory (LSTM) network, and wherein the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> is calculated by adding the products of attention weights for each slot and corresponding hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00002"> <pat:ClaimNumber>2< / pat:ClaimNumber> <pat:ClaimText>2. The method of claim 1, the method further comprising: calculating a hidden state vector hT and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the vectorized original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00003"> <pat:ClaimNumber>3< / pat:ClaimNumber> <pat:ClaimText>3. The method of claim 2, the method further comprising: weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>vI<annotation encoding="application / x-tex">v^I< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00004"> <pat:ClaimNumber>4< / pat:ClaimNumber> <pat:ClaimText>4. The method of claim 3, the method further comprising: employing a slot gate g to join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00005"> <pat:ClaimNumber>5< / pat:ClaimNumber> <pat:ClaimText>5. The method of claim 4, the method further comprising: generating a transformed representation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate g. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00006"> <pat:ClaimNumber>6< / pat:ClaimNumber> <pat:ClaimText>6. The method of claim 5, the method further comprising: jointly optimizing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> to construct a target function. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00007"> <pat:ClaimNumber>7< / pat:ClaimNumber> <pat:ClaimText>7. The method of claim 6, the method further comprising: performing intent recognition on the speech question of the user based on the target function. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00008"> <pat:ClaimNumber>8< / pat:ClaimNumber> <pat:ClaimText>8. The method of claim 7, characterized in that the step of subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process includes: receiving the speech question of the user and transforming the speech question to a recognizable text. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00009"> <pat:ClaimNumber>9< / pat:ClaimNumber> <pat:ClaimText>9. The method of claim 8, characterized in that the step of subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process includes: employing a tokenizer to term-segment the recognizable text and obtain the original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00010"> <pat:ClaimNumber>10< / pat:ClaimNumber> <pat:ClaimText>10. The method of claim 9, characterized in that the step of subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process includes: subjecting the original term sequence to a word embedding process. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00011"> <pat:ClaimNumber>11< / pat:ClaimNumber> <pat:ClaimText>11. The method of claim 10, characterized in that the step of subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process includes: realizing a vector representation of each segmented term in the original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00012"> <pat:ClaimNumber>12< / pat:ClaimNumber> <pat:ClaimText>12. The method of claim 11, characterized in that the step of calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> includes: employing the bidirectional LSTM network to encode each term segmentation vector. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00013"> <pat:ClaimNumber>13< / pat:ClaimNumber> <pat:ClaimText>13. The method of claim 12, characterized in that the step of calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> includes: outputting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> corresponding to each term segmentation vector. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00014"> <pat:ClaimNumber>14< / pat:ClaimNumber> <pat:ClaimText>14. The method of claim 13, characterized in that the step of calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> includes: calculating the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>, to which each term segmentation vector corresponds, through formula <semantics>ciS=∑ai,jShj<annotation encoding="application / x-tex">c_i^S = \sum a_{i,j}^S h_j< / annotation>< / semantics>, wherein <semantics>ai,jS<annotation encoding="application / x-tex">a_{i,j}^S< / annotation>< / semantics> represents an attention weight of a slot, its calculation formula is <semantics>ai,jS=exp(ei,j)∑k=1Texp(ei,k)<annotation encoding="application / x-tex">a_{i,j}^S = \frac{\exp(e_{i,j})}{\sum_{k=1}^T \exp(e_{i,k})}< / annotation>< / semantics>, <semantics>ei,j=σ(WheShj)<annotation encoding="application / x-tex">e_{i,j} = \sigma(W_{he}^S hj)< / annotation>< / semantics>, where <semantics>σ<annotation encoding="application / x-tex">\sigma< / annotation>< / semantics> represents a slot activation function, and <semantics>WheS<annotation encoding="application / x-tex">W_{he}^{S}< / annotation>< / semantics> represents a slot weight matrix. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00015"> <pat:ClaimNumber>15< / pat:ClaimNumber> <pat:ClaimText>15. The method of claim 14, characterized in that the step of calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>viS<annotation encoding="application / x-tex">v_i^S< / annotation>< / semantics> includes: constructing a slot label model <semantics>yiS=softmax(WheS(hi+ciS))<annotation encoding="application / x-tex">y_i^S = softmax (W_{he}^S(hi + c_i^S))< / annotation>< / semantics> based on the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00016"> <pat:ClaimNumber>16< / pat:ClaimNumber> <pat:ClaimText>16. The method of claim 15, characterized in that the step of calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> includes: employing a hidden unit in the bidirectional LSTM network to encode the vectorized original term sequence, and obtaining the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00017"> <pat:ClaimNumber>17< / pat:ClaimNumber> <pat:ClaimText>17. The method of claim 16, characterized in that the step of calculating a hidden state vector hT and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> includes: calculating the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the original term sequence through formula <semantics>cI=∑aiIhT<annotation encoding="application / x-tex">c^I = \sum a_i^I hT< / annotation>< / semantics>, wherein <semantics>aiI<annotation encoding="application / x-tex">a_i^I< / annotation>< / semantics> represents an attention weight of an intent, its calculation formula is <semantics>aiI=<annotation encoding="application / x-tex">a_i^I =< / annotation>< / semantics> <semantics>exp(ej)∑k=1Texp(ek)<annotation encoding="application / x-tex">\frac{\exp(e_j)}{\sum_{k=1}^T \exp(e_k)}< / annotation>< / semantics>, <semantics>ei=σ′(WhvIhT)<annotation encoding="application / x-tex">e_i = \sigma'(W_{hv}^I h T)< / annotation>< / semantics>, where <semantics>σ′<annotation encoding="application / x-tex">\sigma'< / annotation>< / semantics> represents an intent activation function, and <semantics>WhvI<annotation encoding="application / x-tex">W_{hv}^I< / annotation>< / semantics> represents an intent weight matrix. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00018"> <pat:ClaimNumber>18< / pat:ClaimNumber> <pat:ClaimText>18. The method of claim 17, characterized in that the step of calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>vI<annotation encoding="application / x-tex">v^I< / annotation>< / semantics> includes: constructing an intent prediction model <semantics>yI=softmax(WhvI(hT+cI))<annotation encoding="application / x-tex">y^I = softmax(W_{hv}^I(hT + c^I))< / annotation>< / semantics> based on the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00019"> <pat:ClaimNumber>19< / pat:ClaimNumber> <pat:ClaimText>19. The method of claim 18, characterized in that the step of employing a slot gate g to join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>, and generating a transformed representation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate <semantics>g<annotation encoding="application / x-tex">g< / annotation>< / semantics> includes: formally representing the slot gate g as <semantics>g=∑v⋅tanh<annotation encoding="application / x-tex">g = \sum v \cdot tanh< / annotation>< / semantics> <semantics>(cis+W⋅cl)<annotation encoding="application / x-tex">(c_i^s + W \cdot c^l)< / annotation>< / semantics>, wherein v represents a weight vector obtained by training, and W represents a weight matrix obtained by training; and formally representing the transformation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate g [Image disponible dans le document PDF, Image available in the PDF document] < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00020"> <pat:ClaimNumber>20< / pat:ClaimNumber> <pat:ClaimText>20. The method of claim 19, characterized in that the target function constructed by jointly optimizing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>viS<annotation encoding="application / x-tex">v_i^S< / annotation>< / semantics> is: <semantics>p(yS,yI|X)=p(yI|X)∏p(yiS|X)<annotation encoding="application / x-tex">p(y^S, y^I|X) = p(y^I|X) \prod p(y_i^S|X)< / annotation>< / semantics>, wherein <semantics>p(yS,yI|X)<annotation encoding="application / x-tex">p(y^S, y^I|X)< / annotation>< / semantics> represents a conditional probability for outputting slot filling and intent prediction at a given original term sequence, where X is the vectorized original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00021"> <pat:ClaimNumber>21< / pat:ClaimNumber> <pat:ClaimText>21. The method of claim 20, characterized in that the step of performing intent recognition on the speech question of the user based on the target function includes: sequentially obtaining intent conditional probabilities, to which the various segmented terms in the original term sequence correspond, through the target function. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00022"> <pat:ClaimNumber>22< / pat:ClaimNumber> <pat:ClaimText>22. The method of claim 21, characterized in that the step of performing intent recognition on the speech question of the user based on the target function includes: screening therefrom a segmented term with the maximum probability value and recognizing the segmented term as the intent of the speech question of the user. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00023"> <pat:ClaimNumber>23< / pat:ClaimNumber> <pat:ClaimText>23. A human-machine interactive speech recognizing system for an intelligent device, characterized in comprising: a term segmentation processing unit, for subjecting a speech question of a user to a term- segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process; a first calculating unit, for calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics>, wherein the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> is calculated by encoding each term segmentation vector by a bidirectional long short- term memory (LSTM) network, and wherein the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> is calculated by adding the products of attention weights for each slot and corresponding hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00024"> <pat:ClaimNumber>24< / pat:ClaimNumber> <pat:ClaimText>24. The system of claim 23, the system further comprising: a second calculating unit, for calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the vectorized original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00025"> <pat:ClaimNumber>25< / pat:ClaimNumber> <pat:ClaimText>25. The system of claim 24, wherein the second calculating unit is configured to weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> to thereafter obtain an intent prediction <semantics>model yl<annotation encoding="application / x-tex">\text{model } y^l< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00026"> <pat:ClaimNumber>26< / pat:ClaimNumber> <pat:ClaimText>26. The system of claim 25, the system further comprising: a model transforming unit, for employing a slot gate g to join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00027"> <pat:ClaimNumber>27< / pat:ClaimNumber> <pat:ClaimText>27. The system of claim 26, wherein the model transforming unit is configured to generating a transformed representation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate <semantics>g<annotation encoding="application / x-tex">g< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00028"> <pat:ClaimNumber>28< / pat:ClaimNumber> <pat:ClaimText>28. The system of claim 27, the system further comprising: a joint optimization unit, for jointly optimizing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> to construct a target function. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00029"> <pat:ClaimNumber>29< / pat:ClaimNumber> <pat:ClaimText>29. The system of claim 28, wherein the joint optimization unit is configured to performing intent recognition on the speech question of the user based on the target function. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00030"> <pat:ClaimNumber>30< / pat:ClaimNumber> <pat:ClaimText>30. The system of claim 29, characterized in that the term segmentation processing unit includes: a term-segmenting module, for receiving the speech question of the user and transforming the speech question to a recognizable text. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00031"> <pat:ClaimNumber>31< / pat:ClaimNumber> <pat:ClaimText>31. The system of claim 30, wherein the term-segmenting module is configured to employing a tokenizer to term-segment the recognizable text and obtain the original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00032"> <pat:ClaimNumber>32< / pat:ClaimNumber> <pat:ClaimText>32. The system of claim 31, characterized in that the term segmentation processing unit includes an embedding processing module, for subjecting the original term sequence to a word embedding process. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00033"> <pat:ClaimNumber>33< / pat:ClaimNumber> <pat:ClaimText>33. The system of claim 32, wherein the embedding processing module is configured to realizing a vector representation of each segmented term in the original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00034"> <pat:ClaimNumber>34< / pat:ClaimNumber> <pat:ClaimText>34. The system of claim 33, characterized in that the first calculating unit includes: a hidden state calculating module, for employing the bidirectional LSTM network to encode each term segmentation vector, and outputting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> corresponding to each term segmentation vector; a slot context calculating module, for calculating the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>, to which each term segmentation vector corresponds, through formula <semantics>ciS=∑ai,jShj<annotation encoding="application / x-tex">c_i^S = \sum a_{i,j}^S h_j< / annotation>< / semantics>, wherein <semantics>ai,jS<annotation encoding="application / x-tex">a_{i,j}^S< / annotation>< / semantics> represents an attention weight of a slot, its calculation formula is <semantics>ai,jS=exp(ei,j)∑b=1Texp(ei,b)<annotation encoding="application / x-tex">a_{i,j}^S = \frac{\exp(e_{i,j})}{\sum_{b=1}^T \exp(e_{i,b})}< / annotation>< / semantics>, <semantics>ei,j=σ(WheShj)<annotation encoding="application / x-tex">e_{i,j} = \sigma(W_{he}^S hj)< / annotation>< / semantics>, where <semantics>σ<annotation encoding="application / x-tex">\sigma< / annotation>< / semantics> represents a slot activation function, and <semantics>WheS<annotation encoding="application / x-tex">W_{he}^S< / annotation>< / semantics> represents a slot weight matrix; and a slot label model module, for constructing a slot label model <semantics>yiS=<annotation encoding="application / x-tex">y_i^S =< / annotation>< / semantics> softmax <semantics>(WheS(hi+ciS))<annotation encoding="application / x-tex">(W_{he}^{S}(hi+c_{i}^{S}))< / annotation>< / semantics> based on the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_{i}< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^{S}< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00035"> <pat:ClaimNumber>35< / pat:ClaimNumber> <pat:ClaimText>35. The system of claim 34, wherein the hidden state calculating module is configured to outputting the hidden state vector hi corresponding to each term segmentation vector. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00036"> <pat:ClaimNumber>36< / pat:ClaimNumber> <pat:ClaimText>36. The system of claim 35, wherein the first calculating unit includes: a slot context calculating module, for calculating the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>, to which each term segmentation vector corresponds, through formula <semantics>ciS=∑ai,jShj<annotation encoding="application / x-tex">c_i^S = \sum a_{i,j}^S h_j< / annotation>< / semantics>, wherein <semantics>ai,jS<annotation encoding="application / x-tex">a_{i,j}^S< / annotation>< / semantics> represents an attention weight of a slot, its calculation formula is <semantics>ai,jS=exp(ei,j)∑k=1Texp(ei,k)<annotation encoding="application / x-tex">a_{i,j}^S = \frac{\exp(e_{i,j})}{\sum_{k=1}^T \exp(e_{i,k})}< / annotation>< / semantics>, <semantics>ei,j=σ(WheShj)<annotation encoding="application / x-tex">e_{i,j} = \sigma(W_{he}^S hj)< / annotation>< / semantics>, where <semantics>σ<annotation encoding="application / x-tex">\sigma< / annotation>< / semantics> represents a slot activation function, and <semantics>WheS<annotation encoding="application / x-tex">W_{he}^{S}< / annotation>< / semantics> represents a slot weight matrix. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00037"> <pat:ClaimNumber>37< / pat:ClaimNumber> <pat:ClaimText>37. The system of claim 36, wherein the first calculating unit includes: a slot label model module, for constructing a slot label model <semantics>yiS=softmax(WheS(hi+ciS))<annotation encoding="application / x-tex">y_i^S = softmax (W_{he}^S(hi + c_i^S))< / annotation>< / semantics> based on the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00038"> <pat:ClaimNumber>38< / pat:ClaimNumber> <pat:ClaimText>38. The system of claim 37, wherein calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> includes: employing a hidden unit in the bidirectional LSTM network to encode the vectorized original term sequence, and obtaining the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00039"> <pat:ClaimNumber>39< / pat:ClaimNumber> <pat:ClaimText>39. The system of claim 38, wherein calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> includes: calculating the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the original term sequence through formula <semantics>cI=∑aiIhT<annotation encoding="application / x-tex">c^I = \sum a_i^I h T< / annotation>< / semantics>, wherein <semantics>aiI<annotation encoding="application / x-tex">a_i^I< / annotation>< / semantics> represents an attention weight of an intent, its calculation formula is <semantics>aiI<annotation encoding="application / x-tex">a_i^I< / annotation>< / semantics> <semantics>exp(ej)∑k=1Texp(ek)<annotation encoding="application / x-tex">\frac{\exp(e_j)}{\sum_{k=1}^T \exp(e_k)}< / annotation>< / semantics>, <semantics>ei=σ′(WhvIhT)<annotation encoding="application / x-tex">e_i = \sigma'(W_{hv}^I h T)< / annotation>< / semantics>, where <semantics>σ′<annotation encoding="application / x-tex">\sigma'< / annotation>< / semantics> represents an intent activation function, and <semantics>WhvI<annotation encoding="application / x-tex">W_{hv}^I< / annotation>< / semantics> represents an intent weight matrix. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00040"> <pat:ClaimNumber>40< / pat:ClaimNumber> <pat:ClaimText>40. The system of claim 39, wherein calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> includes: constructing an intent prediction model <semantics>yI=softmax(WhvI(hT+cI))<annotation encoding="application / x-tex">y^I = softmax(W_{hv}^I(hT + c^I))< / annotation>< / semantics> based on the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00041"> <pat:ClaimNumber>41< / pat:ClaimNumber> <pat:ClaimText>41. The system of claim 40, wherein employing a slot gate g to join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>, and generating a transformed representation of the slot label model <semantics>viS<annotation encoding="application / x-tex">v_i^S< / annotation>< / semantics> through the slot gate g includes: formally representing the slot gate g as <semantics>g=∑v⋅tanh<annotation encoding="application / x-tex">g = \sum v \cdot tanh< / annotation>< / semantics> <semantics>(cis+W⋅cl)<annotation encoding="application / x-tex">(c_i^s + W \cdot c^l)< / annotation>< / semantics>, wherein v represents a weight vector obtained by training, and W represents a weight matrix obtained by training; and formally representing the transformation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate g as <semantics>yiS=softmax(WheS(hi+ciS⋅g)).<annotation encoding="application / x-tex">y_i^S = softmax(W_{he}^S(hi + c_i^S \cdot g)).< / annotation>< / semantics> < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00042"> <pat:ClaimNumber>42< / pat:ClaimNumber> <pat:ClaimText>42. The system of claim 41, wherein the target function constructed by jointly optimizing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> is: <semantics>p(yS,yI|X)=p(yI|X)∏p(yiS|X)<annotation encoding="application / x-tex">p(y^S, y^I|X) = p(y^I|X) \prod p(y_i^S|X)< / annotation>< / semantics>, wherein <semantics>p(yS,yI|X)<annotation encoding="application / x-tex">p(y^S, y^I|X)< / annotation>< / semantics> represents a conditional probability for outputting slot filling and intent prediction at a given original term sequence, where X is the vectorized original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00043"> <pat:ClaimNumber>43< / pat:ClaimNumber> <pat:ClaimText>43. The system of claim 42, wherein performing intent recognition on the speech question of the user based on the target function includes: sequentially obtaining intent conditional probabilities, to which the various segmented terms in the original term sequence correspond, through the target function. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00044"> <pat:ClaimNumber>44< / pat:ClaimNumber> <pat:ClaimText>44. The system of claim 43, wherein performing intent recognition on the speech question of the user based on the target function includes: screening therefrom a segmented term with the maximum probability value and recognizing the segmented term as the intent of the speech question of the user. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00045"> <pat:ClaimNumber>45< / pat:ClaimNumber> <pat:ClaimText>45. A human-machine interactive speech recognizing device, the equipment comprising: at least one processor; a memory, connected with the at least one processor; wherein the memory stores an instruction executable by the at least one processor configured to: subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process; and calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics>, wherein the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> is calculated by encoding each term segmentation vector by a bidirectional long short-term memory (LSTM) network, and wherein the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> is calculated by adding the products of attention weights for each slot and corresponding hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00046"> <pat:ClaimNumber>46< / pat:ClaimNumber> <pat:ClaimText>46. The device of claim 45, wherein the processor is configured to: calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> of the vectorized original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00047"> <pat:ClaimNumber>47< / pat:ClaimNumber> <pat:ClaimText>47. The device of claim 46, wherein the processor is configured to: weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>vI<annotation encoding="application / x-tex">v^I< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00048"> <pat:ClaimNumber>48< / pat:ClaimNumber> <pat:ClaimText>48. The device of claim 46, wherein the processor is configured to: employing a slot gate g to join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00049"> <pat:ClaimNumber>49< / pat:ClaimNumber> <pat:ClaimText>49. The device of claim 48, wherein the processor is configured to: generating a transformed representation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate g. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00050"> <pat:ClaimNumber>50< / pat:ClaimNumber> <pat:ClaimText>50. The device of claim 49, wherein the processor is configured to: jointly optimizing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> to construct a target function. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00051"> <pat:ClaimNumber>51< / pat:ClaimNumber> <pat:ClaimText>51. The device of claim 50, wherein the processor is configured to: performing intent recognition on the speech question of the user based on the target function. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00052"> <pat:ClaimNumber>52< / pat:ClaimNumber> <pat:ClaimText>52. The device of claim 51, characterized in that the step of subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process includes: receiving the speech question of the user and transforming the speech question to a recognizable text. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00053"> <pat:ClaimNumber>53< / pat:ClaimNumber> <pat:ClaimText>53. The device of claim 52, characterized in that the step of subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process includes: employing a tokenizer to term-segment the recognizable text and obtain the original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00054"> <pat:ClaimNumber>54< / pat:ClaimNumber> <pat:ClaimText>54. The device of claim 53, characterized in that the step of subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process includes: subjecting the original term sequence to a word embedding process. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00055"> <pat:ClaimNumber>55< / pat:ClaimNumber> <pat:ClaimText>55. The device of claim 54, characterized in that the step of subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process includes: realizing a vector representation of each segmented term in the original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00056"> <pat:ClaimNumber>56< / pat:ClaimNumber> <pat:ClaimText>56. The device of claim 55, characterized in that the step of calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> includes: employing the bidirectional LSTM network to encode each term segmentation vector. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00057"> <pat:ClaimNumber>57< / pat:ClaimNumber> <pat:ClaimText>57. The device of claim 56, characterized in that the step of calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> includes: outputting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> corresponding to each term segmentation vector. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00058"> <pat:ClaimNumber>58< / pat:ClaimNumber> <pat:ClaimText>58. The device of claim 57, characterized in that the step of calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> includes: calculating the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>, to which each term segmentation vector corresponds, through formula <semantics>ciS=∑ai,jShi<annotation encoding="application / x-tex">c_i^S = \sum a_{i,j}^S h_i< / annotation>< / semantics>, wherein <semantics>ai,jS<annotation encoding="application / x-tex">a_{i,j}^S< / annotation>< / semantics> represents an attention weight of a slot, its calculation formula is <semantics>ai,jS=exp(ei,j)∑k=1Texp(ei,k)<annotation encoding="application / x-tex">a_{i,j}^S = \frac{\exp(e_{i,j})}{\sum_{k=1}^T \exp(e_{i,k})}< / annotation>< / semantics>, <semantics>ei,j=σ(WheShj)<annotation encoding="application / x-tex">e_{i,j} = \sigma(W_{he}^S hj)< / annotation>< / semantics>, where <semantics>σ<annotation encoding="application / x-tex">\sigma< / annotation>< / semantics> represents a slot activation function, and <semantics>WheS<annotation encoding="application / x-tex">W_{he}^{S}< / annotation>< / semantics> represents a slot weight matrix. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00059"> <pat:ClaimNumber>59< / pat:ClaimNumber> <pat:ClaimText>59. The device of claim 58, characterized in that the step of calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> includes: constructing a slot label model <semantics>yiS=softmax(WheS(hi+ciS))<annotation encoding="application / x-tex">y_i^S = softmax (W_{he}^S(hi + c_i^S))< / annotation>< / semantics> based on the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00060"> <pat:ClaimNumber>60< / pat:ClaimNumber> <pat:ClaimText>60. The device of claim 59, characterized in that the step of calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>vI<annotation encoding="application / x-tex">v^I< / annotation>< / semantics> includes: employing a hidden unit in the bidirectional LSTM network to encode the vectorized original term sequence, and obtaining the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00061"> <pat:ClaimNumber>61< / pat:ClaimNumber> <pat:ClaimText>61. The device of claim 60, characterized in that the step of calculating a hidden state vector hT and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> includes: calculating the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the original term sequence through formula <semantics>cI=∑aiIhT<annotation encoding="application / x-tex">c^I = \sum a_i^I h T< / annotation>< / semantics>, wherein <semantics>aiI<annotation encoding="application / x-tex">a_i^I< / annotation>< / semantics> represents an attention weight of an intent, its calculation formula is <semantics>aiI<annotation encoding="application / x-tex">a_i^I< / annotation>< / semantics> = <semantics>exp(ej)∑k=1Texp(ek)<annotation encoding="application / x-tex">\frac{\exp(e_j)}{\sum_{k=1}^T \exp(e_k)}< / annotation>< / semantics>, <semantics>ei=σ′(WhvIhT)<annotation encoding="application / x-tex">e_i = \sigma'(W_{hv}^I h T)< / annotation>< / semantics>, where <semantics>σ′<annotation encoding="application / x-tex">\sigma'< / annotation>< / semantics> represents an intent activation function, and <semantics>WhvI<annotation encoding="application / x-tex">W_{hv}^I< / annotation>< / semantics> represents an intent weight matrix. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00062"> <pat:ClaimNumber>62< / pat:ClaimNumber> <pat:ClaimText>62. The device of claim 61, characterized in that the step of calculating a hidden state vector hT and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>vI<annotation encoding="application / x-tex">v^I< / annotation>< / semantics> includes: constructing an intent prediction model <semantics>yI=softmax(WhvI(hT+cI))<annotation encoding="application / x-tex">y^I = softmax(W_{hv}^I(hT + c^I))< / annotation>< / semantics> based on the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00063"> <pat:ClaimNumber>63< / pat:ClaimNumber> <pat:ClaimText>63. The device of claim 62, characterized in that the step of employing a slot gate g to join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>, and generating a transformed representation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate g includes: formally representing the slot gate g as <semantics>g=∑v⋅tanh<annotation encoding="application / x-tex">g = \sum v \cdot tanh< / annotation>< / semantics> (<semantics>cis+W⋅cl<annotation encoding="application / x-tex">c_i^s + W \cdot c^l< / annotation>< / semantics>), wherein v represents a weight vector obtained by training, and W represents a weight matrix obtained by training; and formally representing the transformation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate g [Image disponible dans le document PDF, Image available in the PDF document] < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00064"> <pat:ClaimNumber>64< / pat:ClaimNumber> <pat:ClaimText>64. The device of claim 63, characterized in that the target function constructed by jointly optimizing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>viS<annotation encoding="application / x-tex">v_i^S< / annotation>< / semantics> is: <semantics>p(yS,yI|X)=p(yI|X)∏p(yiS|X)<annotation encoding="application / x-tex">p(y^S, y^I | X) = p(y^I | X) \prod p(y_i^S | X)< / annotation>< / semantics>, wherein <semantics>p(yS,yI|X)<annotation encoding="application / x-tex">p(y^S, y^I | X)< / annotation>< / semantics> represents a conditional probability for outputting slot filling and intent prediction at a given original term sequence, where X is the vectorized original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00065"> <pat:ClaimNumber>65< / pat:ClaimNumber> <pat:ClaimText>65. The device of claim 64, characterized in that the step of performing intent recognition on the speech question of the user based on the target function includes: sequentially obtaining intent conditional probabilities, to which the various segmented terms in the original term sequence correspond, through the target function. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00066"> <pat:ClaimNumber>66< / pat:ClaimNumber> <pat:ClaimText>66. The device of claim 65, characterized in that the step of performing intent recognition on the speech question of the user based on the target function includes: screening therefrom a segmented term with the maximum probability value and recognizing the segmented term as the intent of the speech question of the user. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00067"> <pat:ClaimNumber>67< / pat:ClaimNumber> <pat:ClaimText>67. A computer-readable storage medium, storing a computer-executable instructions thereon, when executed by a computer, the computer is configured to: subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process; and calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics>, wherein the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> is calculated by encoding each term segmentation vector by a bidirectional long short-term memory (LSTM) network, and wherein the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> is calculated by adding the products of attention weights for each slot and corresponding hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00068"> <pat:ClaimNumber>68< / pat:ClaimNumber> <pat:ClaimText>68. The storage medium of claim 67, the storage medium further comprising: calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the vectorized original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00069"> <pat:ClaimNumber>69< / pat:ClaimNumber> <pat:ClaimText>69. The storage medium of claim 68, the storage medium further comprising: weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00070"> <pat:ClaimNumber>70< / pat:ClaimNumber> <pat:ClaimText>70. The storage medium of claim 69, the storage medium further comprising: employing a slot gate g to join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00071"> <pat:ClaimNumber>71< / pat:ClaimNumber> <pat:ClaimText>71. The storage medium of claim 70, the storage medium further comprising: generating a transformed representation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate <semantics>g<annotation encoding="application / x-tex">g< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00072"> <pat:ClaimNumber>72< / pat:ClaimNumber> <pat:ClaimText>72. The storage medium of claim 71, the storage medium further comprising: jointly optimizing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> to construct a target function. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00073"> <pat:ClaimNumber>73< / pat:ClaimNumber> <pat:ClaimText>73. The storage medium of claim 72, the storage medium further comprising: performing intent recognition on the speech question of the user based on the target function. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00074"> <pat:ClaimNumber>74< / pat:ClaimNumber> <pat:ClaimText>74. The storage medium of claim 73, characterized in that the step of subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process includes: receiving the speech question of the user and transforming the speech question to a recognizable text. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00075"> <pat:ClaimNumber>75< / pat:ClaimNumber> <pat:ClaimText>75. The storage medium of claim 74, characterized in that the step of subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process includes: employing a tokenizer to term- segment the recognizable text and obtain the original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00076"> <pat:ClaimNumber>76< / pat:ClaimNumber> <pat:ClaimText>76. The storage medium of claim 75, characterized in that the step of subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process includes: subjecting the original term sequence to a word embedding process. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00077"> <pat:ClaimNumber>77< / pat:ClaimNumber> <pat:ClaimText>77. The storage medium of claim 76, characterized in that the step of subjecting a speech question of a user to a term-segmenting process to obtain an original term sequence, and vectorizing the original term sequence through an embedding process includes: realizing a vector representation of each segmented term in the original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00078"> <pat:ClaimNumber>78< / pat:ClaimNumber> <pat:ClaimText>78. The storage medium of claim 77, characterized in that the step of calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> includes: employing the bidirectional LSTM network to encode each term segmentation vector. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00079"> <pat:ClaimNumber>79< / pat:ClaimNumber> <pat:ClaimText>79. The storage medium of claim 78, characterized in that the step of calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> includes: outputting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> corresponding to each term segmentation vector. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00080"> <pat:ClaimNumber>80< / pat:ClaimNumber> <pat:ClaimText>80. The storage medium of claim 79, characterized in that the step of calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> includes: calculating the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>, to which each term segmentation vector corresponds, through formula <semantics>ciS=∑ai,jShj<annotation encoding="application / x-tex">c_i^S = \sum a_{i,j}^S h_j< / annotation>< / semantics>, wherein <semantics>ai,jS<annotation encoding="application / x-tex">a_{i,j}^S< / annotation>< / semantics> represents an attention weight of a slot, its calculation formula is <semantics>ai,jS=exp(ei,j)∑k=1Texp(ei,k)<annotation encoding="application / x-tex">a_{i,j}^S = \frac{\exp(e_{i,j})}{\sum_{k=1}^T \exp(e_{i,k})}< / annotation>< / semantics>, <semantics>ei,j=σ(WheShj)<annotation encoding="application / x-tex">e_{i,j} = \sigma(W_{he}^S hj)< / annotation>< / semantics>, where <semantics>σ<annotation encoding="application / x-tex">\sigma< / annotation>< / semantics> represents a slot activation function, and <semantics>WheS<annotation encoding="application / x-tex">W_{he}^{S}< / annotation>< / semantics> represents a slot weight matrix. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00081"> <pat:ClaimNumber>81< / pat:ClaimNumber> <pat:ClaimText>81. The storage medium of claim 80, characterized in that the step of calculating a hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and a slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> of each term segmentation vector, and weighting the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> to thereafter obtain a slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> includes: constructing a slot label model <semantics>yiS=softmax(WheS(hi+ciS))<annotation encoding="application / x-tex">y_i^S = softmax (W_{he}^S(hi + c_i^S))< / annotation>< / semantics> based on the hidden state vector <semantics>hi<annotation encoding="application / x-tex">h_i< / annotation>< / semantics> and the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00082"> <pat:ClaimNumber>82< / pat:ClaimNumber> <pat:ClaimText>82. The storage medium of claim 81, characterized in that the step of calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> includes: employing a hidden unit in the bidirectional LSTM network to encode the vectorized original term sequence, and obtaining the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00083"> <pat:ClaimNumber>83< / pat:ClaimNumber> <pat:ClaimText>83. The storage medium of claim 82, characterized in that the step of calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> includes: calculating the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the original term sequence through formula <semantics>cI=∑aiIhT<annotation encoding="application / x-tex">c^I = \sum a_i^I h T< / annotation>< / semantics>, wherein <semantics>aiI<annotation encoding="application / x-tex">a_i^I< / annotation>< / semantics> represents an attention weight of an intent, its calculation formula is <semantics>aiI=<annotation encoding="application / x-tex">a_i^I =< / annotation>< / semantics> <semantics>exp(ej)∑k=1Texp(ek)<annotation encoding="application / x-tex">\frac{\exp(e_j)}{\sum_{k=1}^T \exp(e_k)}< / annotation>< / semantics>, <semantics>ei=σ′(WhvIhT)<annotation encoding="application / x-tex">e_i = \sigma'(W_{hv}^I h T)< / annotation>< / semantics>, where <semantics>σ′<annotation encoding="application / x-tex">\sigma'< / annotation>< / semantics> represents an intent activation function, and <semantics>WhvI<annotation encoding="application / x-tex">W_{hv}^I< / annotation>< / semantics> represents an intent weight matrix. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00084"> <pat:ClaimNumber>84< / pat:ClaimNumber> <pat:ClaimText>84. The storage medium of claim 83, characterized in that the step of calculating a hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and an intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics> of the vectorized original term sequence, and weighting the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^{I}< / annotation>< / semantics> to thereafter obtain an intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> includes: constructing an intent prediction model <semantics>yI=softmax(WhvI(hT+cI))<annotation encoding="application / x-tex">y^I = softmax(W_{hv}^I(hT + c^I))< / annotation>< / semantics> based on the hidden state vector <semantics>hT<annotation encoding="application / x-tex">hT< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00085"> <pat:ClaimNumber>85< / pat:ClaimNumber> <pat:ClaimText>85. The storage medium of claim 84, characterized in that the step of employing a slot gate g to join the slot context vector <semantics>ciS<annotation encoding="application / x-tex">c_i^S< / annotation>< / semantics> and the intent context vector <semantics>cI<annotation encoding="application / x-tex">c^I< / annotation>< / semantics>, and generating a transformed representation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate <semantics>g<annotation encoding="application / x-tex">g< / annotation>< / semantics> includes: formally representing the slot gate g as <semantics>g=∑v⋅tanh<annotation encoding="application / x-tex">g = \sum v \cdot tanh< / annotation>< / semantics> <semantics>(cis+W⋅cl)<annotation encoding="application / x-tex">(c_i^s + W \cdot c^l)< / annotation>< / semantics>, wherein v represents a weight vector obtained by training, and W represents a weight matrix obtained by training; and formally representing the transformation of the slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> through the slot gate g as <semantics>yiS=softmax(WheS(hi+ciS⋅g)).<annotation encoding="application / x-tex">y_i^S = softmax(W_{he}^S(hi + c_i^S \cdot g)).< / annotation>< / semantics> < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00086"> <pat:ClaimNumber>86< / pat:ClaimNumber> <pat:ClaimText>86. The storage medium of claim 85, characterized in that the target function constructed by jointly optimizing the intent prediction model <semantics>yI<annotation encoding="application / x-tex">y^I< / annotation>< / semantics> and the transformed slot label model <semantics>yiS<annotation encoding="application / x-tex">y_i^S< / annotation>< / semantics> is: <semantics>p(yS,yI|X)=p(yI|X)∏p(yiS|X)<annotation encoding="application / x-tex">p(y^S, y^I | X) = p(y^I | X) \prod p(y_i^S | X)< / annotation>< / semantics>, wherein <semantics>p(yS,yI|X)<annotation encoding="application / x-tex">p(y^S, y^I | X)< / annotation>< / semantics> represents a conditional probability for outputting slot filling and intent prediction at a given original term sequence, where X is the vectorized original term sequence. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00087"> <pat:ClaimNumber>87< / pat:ClaimNumber> <pat:ClaimText>87. The storage medium of claim 86, characterized in that the step of performing intent recognition on the speech question of the user based on the target function includes: sequentially obtaining intent conditional probabilities, to which the various segmented terms in the original term sequence correspond, through the target function. < / pat:ClaimText> < / pat:Claim> <pat:Claim com:id="CLM-00088"> <pat:ClaimNumber>88< / pat:ClaimNumber> <pat:ClaimText>88. The storage medium of claim 87, characterized in that the step of performing intent recognition on the speech question of the user based on the target function includes: screening therefrom a segmented term with the maximum probability value and recognizing the segmented term as the intent of the speech question of the user. < / pat:ClaimText> < / pat:Claim> < / pat:Claims>