CN115662392A - Transliteration method based on phoneme memory, electronic equipment and storage medium - Google Patents
Transliteration method based on phoneme memory, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN115662392A CN115662392A CN202211595293.3A CN202211595293A CN115662392A CN 115662392 A CN115662392 A CN 115662392A CN 202211595293 A CN202211595293 A CN 202211595293A CN 115662392 A CN115662392 A CN 115662392A
- Authority
- CN
- China
- Prior art keywords
- phoneme
- letter
- layer
- vector
- formula
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Machine Translation (AREA)
Abstract
The invention discloses a transliteration method based on phoneme memory, electronic equipment and a storage medium, wherein the transliteration method comprises the following steps: 1. extracting the transliterated words and splitting the transliterated words into letters, 2, constructing a phoneme library, and extracting phoneme features associated with each letter; 3. constructing an encoder of an L layer, and encoding letters to obtain letter encoding vectors corresponding to each letter in each layer; 4. establishing an L-layer phoneme memory network for modeling letter coding vectors and phoneme characteristics to obtain a letter coding matrix; 5. inputting the letter coding matrix and a target letter output by the classifier at the previous t moment into a decoder of an L layer, and sending an obtained letter prediction vector output by the decoder at the t moment into the classifier to obtain a predicted target letter at the t moment; 6. assigning T +1 to T, and repeating the step 5 until the time T, thereby obtaining the predicted letter sequence. The invention aims to fuse the phoneme characteristics into a standard text generation process, thereby improving the transliteration quality and effect.
Description
Technical Field
The invention belongs to the field of natural language processing, and particularly relates to a transliteration method based on phoneme memory, electronic equipment and a storage medium.
Background
Transliteration refers to translating a person's name in a source language, such as Smith, to text in a target language, such as chinese, such as Smith) without changing the pronunciation of the name in the source language. For example, the name "Smith" in the Source language English is transliterated to "Smith" in Chinese.
The existing method mostly regards the task as a sequence-to-sequence generation task, and adopts a high-level encoder and a decoder to generate name transliteration of a target language, and lacks the utilization of phonetic features, especially phoneme features, in a source language and a target language, so that words generated by transliteration lose pronunciation features of the source language, and the accuracy of transliteration is reduced.
Disclosure of Invention
The present invention is directed to overcome the above-mentioned shortcomings of the prior art, and provides a transliteration method, an electronic device and a storage medium based on phoneme memory, so as to integrate phoneme features into a standard text generation process, thereby improving the transliteration quality and effect.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention relates to a transliteration method based on phoneme memory, which is characterized by comprising the following steps:
step 1, extracting a plurality of transliterated words from a source language corpus, and splitting each word into letters; wherein, the ith word X i The split letter sequence is recorded asx i,1 ,…x i,j ,…,},x i,j Represents the ith word X i The jth letter in (1), n i Represents the ith word X i The total number of middle letters;
step 2, selecting the jth letter from the phoneme libraryx i,j Associated m phoneme characteristics and forming a phoneme characteristic setS i,j ={s i,j,1 …s i,j,u ,…s i,j,m And (c) the step of (c) in which,s i,j,u is the j-th letterx i,j An associated u-th phoneme feature, m being a total number of associated phoneme features;
step 3, constructing a transliteration network, comprising: an L-layer encoder, an L-layer phoneme memory network, an L-layer decoder and a classifier;
step 3.1, processing of an encoder:
the j letterx i,j Conversion to jth letter vectorThen inputting the encoded vector into the encoder, sequentially processing the encoded vector by a multi-head self-attention layer of L layers to obtain L letter encoded vectors|l=1,2, \8230;, L }; wherein the content of the first and second substances,is shown aslThe multi-head of the layer outputs the jth letter encoding vector from the attention layer;
step 3.2, processing the phoneme memory network:
collecting phoneme characteristicsS i,j Conversion into phoneme vector set| u =1,2, \8230 |, m } back, and a hard page|l=1,2, \ 8230;, L } are input into the phoneme memory network together for processing, and n is enhanced i Letter code vector|l=1,2,…,L;j=1,2,…,n i And is marked as the ith word X i Alpha-code matrix H i (ii) a Wherein the content of the first and second substances,representing the u-th phonemes i,j,u The phoneme vector of (a);representing an enhanced jth letter code vector;
and 3.3, processing by a decoder:
encoding letters into matrix H i Inputting the letter and the target letter output by the classifier at the previous t moment into a decoder of an L layer, and obtaining a letter prediction vector h output by the decoder at the t moment i,t (ii) a When t =1, making the letter output by the classifier at the previous t moment be null;
and 3.4, processing by a classifier:
the classifier utilizes the full-connection layer to predict the letter vector h output by the decoder at the time t i,t Processing to obtain the ith word X at the current time t i Predicted target letter y i,t ;
Step 3.5, after assigning t +1 to t, returning to the step 3.3 for sequential execution until the step is finishedTUntil the moment, thereby obtaining the ith word X i Predicted letter sequence of { y } i,1 ,…, y i,t ,…, y i,T }。
The transliteration method based on phoneme memory of the invention is also characterized in that the step 2 comprises the following steps:
step 2.1, calculate the jth letter using equation (1)x i,j With the qth phoneme feature in the phoneme librarys q Point-by-point mutual information PMI of (x i,j ,s q ) To obtain the j letterx i,j Point-to-point mutual information with all M phoneme features { PMI (PMI) ((PMI))x i,j ,s q )|1<=q<=M };MRepresenting the number of all phoneme characteristics in the phoneme library;
in the formula (1), p: (x i,j ,s q ) Represents the j letterx i,j With the qth phoneme features q Probability of co-occurrence; p (a)x i,j ) Represents the j letterx i,j Appearing in the ith word X i A probability of (1); p (a)s q ) Representing the qth phoneme features q Appearing in the ith word X i Probability in pronunciation of (a);
step 2.2 from point-to-point mutual information { PMI ((PMI))x i,j ,s q )|1<=q<Selecting M phoneme characteristics corresponding to the highest point-by-point mutual information from the M phoneme characteristics, and forming a phoneme characteristic setS i,j ={s i,j,1 …s i,j,u ,…s i,j,m }。
Said step 3.2 comprises:
step 3.2.1, the u-th phonemes i,j,u Conversion to the u-th phoneme vectorThen, withAre input together intolIn a hierarchical phoneme memory network, said firstlThe layer phoneme memory network utilizes the pairs of formula (2) and formula (3)After mapping, get the secondlU-th phoneme key vector of layerAnd a firstlU-th phoneme value vector of layer:
In the formula (1) and the formula (2),is shown aslThe key matrix of the layer(s),denotes the firstlA matrix of values for the layers; reLU denotes the activation function; "·" denotes the multiplication of a matrix and a vector;
step 3.2.2, saidlThe phoneme memory network of the layer calculates the second by using the formula (4)lU-th phoneme weight of layer:
In the formula (3), ". Denotes a vector inner product;
step 3.2.3, saidlThe hierarchical phoneme memory network calculates a weighted average vector using equation (5):
Step 3.2.4, saidlThe phoneme memory network of the layer is obtained by using the formula (6)lLayer jth letter reset vector:
In the formula (5), sigmoid represents an activation function,andrespectively representlA first reset matrix and a second reset matrix of a layer,is shown aslA reset offset vector of a layer;
step 3.2.5, saidlThe phoneme memory network of the layer is obtained by using the formula (7)lLayer enhanced jth letter code vectorSo that the enhanced jth letter code vector is output by the L-layer phoneme memory network|l=1,2, \ 8230;, L }, and thus enhanced n i Letter code vector|l=1,2,…,L, j=1,2,…,n i And is marked as the ith word X i Alpha-code matrix H i ;
In the formula (7), the reaction mixture is,the product of the hadamard is represented,representing a vector concatenation, with 1 representing a vector with all dimension values of 1.
The electronic device of the invention comprises a memory and a processor, and is characterized in that the memory is used for storing programs for supporting the processor to execute the transliteration method, and the processor is configured to execute the programs stored in the memory.
The invention relates to a computer-readable storage medium, on which a computer program is stored, characterized in that the computer program executes the steps of the transliteration method when being executed by a processor.
Compared with the prior art, the invention has the beneficial effects that:
1. according to the method, for each letter in the input, the representation of the letter is enhanced by using the phoneme characteristics associated with the letter through the phoneme memory neural network of the L layer, so that the understanding of the model to the pronunciation characteristics of the target language is enhanced, the transliteration text generated by the model retains the phonetic characteristics of the source language as far as possible, and the model transliteration performance is improved.
2. The method realizes the identification and utilization of the importance of different phoneme characteristics by weighting different phoneme characteristics, and effectively avoids the influence of potential noise in the phoneme characteristics on the model performance.
Drawings
FIG. 1 is a flow chart of the transliteration method of the present invention.
Detailed Description
In this embodiment, a transliteration method based on phoneme memory is performed as shown in fig. 1, and includes the following steps:
step 1, extracting a plurality of transliterated words from a source language corpus, and splitting each word into letters; wherein, the ith word X i The split letter sequence is recorded asx i,1 ,…x i,j ,…, },x i,j Represents the ith word X i The j-th letter in (1), n i Represents the ith word X i The total number of middle letters; for example, 4 transliterated words extracted from an english source language corpus are { Tom, smith, bob, cook }, and then after the words are split, the letter sequence of the split 2 nd word "Smith" is { "S", "m", "i", "t", "h", wherein there are 5 letters in total, and the 3 rd letter is "i".
Step 2, selecting the jth letter from the phoneme libraryx i,j Associated m phoneme characteristics and forming a phoneme characteristic setS i,j ={s i,j,1 …s i,j,u ,…s i,j,m And (c) the step of (c) in which,s i,j,u is the jth letterx i,j An associated u-th phoneme feature, m being a total number of associated phoneme features; for example, the phoneme library is a set including all international phonetic symbols { "a", "e", "o", "t", "g", "k", "i", "i:", "is" I ", \8230;". When m =3, the phoneme feature set extracted from it and associated with the 3 rd letter "i" is { "i:", { "i:" I "," i "}. The 1 st phoneme feature associated with the 3 rd letter is "i:".
Step 2.1, calculate the jth letter using equation (1)x i,j And the q-th phoneme feature in the phoneme librarys q PMI (a) of point-by-point mutual informationx i,j ,s q ) To obtain the j letterx i,j Point-to-point mutual information with all M phoneme features { PMI (PMI) ((PMI))x i,j ,s q )|1<=q<=M };MRepresenting the number of all phoneme characteristics in the phoneme library;
in the formula (1), p: (x i,j ,s q ) Represents the j letterx i,j With the qth phoneme features q Probability of co-occurrence; p (a)x i,j ) Represents the j letterx i,j Appearing in the ith word X i The probability of (2); p (a)s q ) Representing the qth phoneme features q Appearing in the ith word X i Is determined. For example, the process of calculating the 3 rd letter "i" and the 8 th phoneme feature "i:" in the phoneme library is. The probability 0.6 of co-occurrence of the 3 rd letter "i" with the 8 th phoneme feature "i:" in the phoneme library is calculated, the probability 0.3 of the 3 rd letter "i" occurring in the 2 nd word "Smith" is calculated, and the probability 0.5 of the 8 th phoneme feature "i:" occurring in the pronunciation of "Smith" is calculated. Using formula (1), the point-to-point mutual information of the 3 rd letter "i" and the 8 th phoneme feature "i:" in the phoneme library is calculated to be 2. By the same method, the point-by-point mutual information of the 3 rd letter "i" and all the phoneme characteristics in the phoneme library can be calculated.
Step 2.2 from point-to-point mutual information { PMI ((PMI))x i,j ,s q )|1<=q<Selecting M phoneme characteristics corresponding to the highest point-by-point mutual information from the = M } and forming a phoneme characteristic setS i,j ={s i,j,1 …s i,j,u ,…s i,j,m }. For example, the 3 highest-scoring phoneme feature of the 3 rd letter "i" is { "i:", { "i:" I ”, “i”}。
Step 3, building a transliteration network, comprising the following steps: an L-layer encoder, an L-layer phoneme memory network, an L-layer decoder and a classifier;
and 3.1, processing of an encoder:
the j letterx i,j Conversion to jth letter vectorThen inputting the encoded vector into the encoder, sequentially processing the encoded vector by a multi-head self-attention layer of L layers to obtain L letter encoded vectors|l=1,2, \8230;, L }; wherein, the first and the second end of the pipe are connected with each other,denotes the firstlThe multi-head of the layer outputs the jth letter encoding vector from the attention layer; for example, when L =6, the 3 rd letter "i" is first converted into an alphabet vector, and then, after processing of a multi-head self-attention layer of 6 layers, a 6-letter coded vector of the 3 rd letter "i" is obtained.
Step 3.2, processing the phoneme memory network:
collecting phoneme characteristicsS i,j Converting into phoneme vector set| u =1,2, \8230 |, m } back, and a hard page|l=1,2, \8230:, L is input into the phoneme memory network together for processing, and n is enhanced i Letter code vector|l=1,2,…,L;j=1,2,…,n i And is marked as the ith word X i Alpha-code matrix H i (ii) a Wherein the content of the first and second substances,representing the u-th phonemes i,j,u The phoneme vector of (a);representing an enhanced jth letter code vector; for example, the phoneme feature set of the 3 rd letter "i {" i: ", {" i: " I "," i "} is converted to a set of 3 phoneme vectors, which 3 phoneme vectors are input into the phoneme memory network together with the 6 letter code vectors for the 3 rd letter" i ", resulting in an enhanced 3 rd letter code vectorThen, the same operation is carried out on all the letters to obtain a letter coding matrix H of the 2 nd word "Smith 2 。
Step 3.2.1, the u-th phoneme features i,j,u Conversion to the u-th phoneme vectorThen, withAre input together intolIn a hierarchical phoneme memory network, said firstlThe layer phoneme memory network utilizes the pairs of formula (2) and formula (3)After mapping, the first one is obtainedlLayer u-th phoneme key vectorAnd a firstlU-th phoneme value vector of layer:
In the formula (1) and the formula (2),is shown aslThe key matrix of the layer(s),denotes the firstlA matrix of values for the layers; reLU denotes activation function; "·" denotes the multiplication of a matrix and a vector; for example, the 1 st phoneme feature "i:" is converted into a 1 st phoneme vector, and the 1 st phoneme vector and the 3 rd letter "i" are input into the phoneme memory network of the 4 th layer, so as to obtain the 1 st phoneme key vector of the 4 th layer and the first phoneme value vector of the 4 th layer.
Step 3.2.2, saidlThe phoneme memory network of the layer is calculated by using the formula (4)lU-th phoneme weight of layer:
In the formula (3), ". Denotes a vector inner product; for example, the 3 rd letter "i" has a total of three phoneme weights at level 4, the first of which is 0.5, the second of which is 0.3, and the third of which is 0.2.
Step 3.2.3, saidlThe hierarchical phoneme memory network calculates a weighted average vector using equation (5):
For example, the weighted average vector of the layer 4 of the 3 rd letter "i" is the weighted average of the value vectors of the layer 4 of the 3 rd letter "i", which in turn has a weight of 0.5,0.3,0.2.
Step 3.2.4, saidlPhoneme memory of layersThe network utilizes formula (6) to obtainlLayer jth letter reset vector:
In the formula (5), sigmoid represents an activation function,andrespectively representlA first reset matrix and a second reset matrix of a layer,is shown aslReset offset vector of layer. Layer i jth letter reset vector due to the nature of sigmoid activation functionThe value of each dimension is between 0-1, representing the reset weight for each dimension of the vector.
Step 3.2.5, saidlThe phoneme memory network of the layer is obtained by using the formula (7)lLayer enhanced jth letter coded vectorSo that the enhanced jth letter code vector is output from the phoneme memory network of L layer|l=1,2, \ 8230;, L }, and thus enhanced n i Letter code vector|l=1,2,…,L, j=1,2,…,n i And is marked as the ith word X i Alphabet coding matrix H of i ;
In the formula (7), the reaction mixture is,the hadamard product is represented by,representing a vector concatenation, with 1 representing a vector with all dimension values of 1.Andplay a role inlJth letter coded vector of a layerAnd a first step oflJth weighted average vector of a layerThe contribution of each dimension is weighted separately.
And 3.3, processing by a decoder:
encoding letters into matrix H i Inputting the letter and the target letter output by the classifier at the previous t moment into a decoder of an L layer, and obtaining a letter prediction vector h output by the decoder at the t moment i,t (ii) a When t =1, making the letter output by the classifier at the previous t moment be null; for example, when t =1, the input to the decoder is the letter code matrix H i And a special letter for indicating an empty letter { "<s>"}. When t =3, the input of the decoder is the letter code matrix H i And the target letters { "history", "secret" } that the classifier has output.
And 3.4, processing by a classifier:
the classifier using full-link layers for the output of the decoder at time tLetter prediction vector h i,t Processing to obtain the ith word X at the current time t i Predicted target letter y i,t (ii) a For example, when t =1, the target letter predicted for the 2 nd word "Smith" at time t is "history"; when t =3 is, the target letter predicted for the 2 nd word "Smith" at time t is "s".
Step 3.5, after assigning t +1 to t, returning to the step 3.3 for sequential execution until the step is finishedTUntil the moment, thereby obtaining the ith word X i Predicted letter sequence y i,1 ,…, y i,t ,…, y i,T }. Specific judgment timeTIs that the ith word X is at time T +1 i Is a predictive letter of "</s>". For example, the 2 nd word X at time t =4 i Is predicted to be "</s>", then T =3, the predicted letter sequence {" history "," dense "," si "} for the 2 nd word" Smith "can be obtained.
In this embodiment, an electronic device includes a memory for storing a program that supports a processor to execute the transliteration method described above, and a processor configured to execute the program stored in the memory.
In this embodiment, a computer-readable storage medium stores a computer program, and the computer program is executed by a processor to perform the steps of the transliteration method.
Claims (5)
1. A transliteration method based on phoneme memory is characterized by comprising the following steps:
step 1, extracting a plurality of transliterated words from a source language corpus, and splitting each word into letters; wherein, the ith word X i The split letter sequence is recorded asx i,1 ,…x i,j ,…,},x i,j Represents the ith word X i The j-th letter in (1), n i Represents the ith word X i The total number of middle letters;
step 2, selecting the jth letter from the phoneme libraryx i,j M associated phoneme characteristics and forming a phoneme characteristic setS i,j ={s i,j,1 …s i,j,u ,…s i,j,m And (c) the step of (c) in which,s i,j,u is the jth letterx i,j An associated u-th phoneme feature, m being a total number of associated phoneme features;
step 3, building a transliteration network, comprising the following steps: an L-layer encoder, an L-layer phoneme memory network, an L-layer decoder and a classifier;
step 3.1, processing of an encoder:
the jth letterx i,j Conversion to jth letter vectorThen inputting the encoded vector into the encoder, sequentially processing the encoded vector by a multi-head self-attention layer of L layers to obtain L letter encoded vectors|l=1,2, \8230;, L }; wherein the content of the first and second substances,denotes the firstlThe multi-head of the layer outputs the jth letter encoding vector from the attention layer;
step 3.2, processing the phoneme memory network:
collecting phoneme feature setS i,j Conversion into phoneme vector set| u =1,2, \8230;, m } rear, with a great opening|l=1,2, \ 8230;, L } are input together into the phoneme memory netProcessing in the net to obtain enhanced n i Letter code vector|l=1,2,…,L;j=1,2,…,n i And is marked as the ith word X i Alpha-code matrix H i (ii) a Wherein the content of the first and second substances,representing the u-th phonemes i,j,u The phoneme vector of (a);representing an enhanced jth letter code vector;
and 3.3, processing by a decoder:
encoding letters into matrix H i Inputting the letter and the target letter output by the classifier at the previous t moment into a decoder of an L layer, and obtaining a letter prediction vector h output by the decoder at the t moment i,t (ii) a When t =1, making the letter output by the classifier at the previous t moment be null;
and 3.4, processing by a classifier:
the classifier utilizes the full-connection layer to predict the letter vector h output by the decoder at the time t i,t Processing to obtain the ith word X at the current time t i Predicted target letter y i,t ;
Step 3.5, after assigning t +1 to t, returning to the step 3.3 for sequential execution until the step is finishedTUntil the moment, thereby obtaining the ith word X i Predicted letter sequence of { y } i,1 ,…, y i,t ,…, y i,T }。
2. The transliteration method based on phoneme memory as claimed in claim 1, wherein said step 2 comprises:
step 2.1, calculate the jth letter using equation (1)x i,j And the q-th phoneme feature in the phoneme librarys q Point-by-point mutual information PMI of (x i,j ,s q ) To obtain the j letterx i,j Point-to-point mutual information with all M phoneme features { PMI: (x i,j ,s q )|1<=q<=M };MRepresenting the number of all phoneme characteristics in the phoneme library;
in the formula (1), p: (1)x i,j ,s q ) Represents the j letterx i,j With the qth phoneme features q Probability of co-occurrence; p (a)x i,j ) Represents the jth letterx i,j Appearing in the ith word X i The probability of (2); p (a)s q ) Representing the qth phoneme features q Appearing in the ith word X i Probability in pronunciation of (a);
step 2.2 from point-to-point mutual information { PMI ((PMI))x i,j ,s q )|1<=q<Selecting M phoneme characteristics corresponding to the highest point-by-point mutual information from the M phoneme characteristics, and forming a phoneme characteristic setS i,j ={s i,j,1 …s i,j,u ,…s i,j,m }。
3. Transliteration method based on phoneme memory according to claim 1, characterised in that said step 3.2 comprises:
step 3.2.1, the u-th phonemes i,j,u Conversion to the u-th phoneme vectorThen, withAre input together intolIn a hierarchical phoneme memory network, the firstlLayer phoneme memory network using type (2)And formula (3) pairAfter mapping, the first one is obtainedlLayer u-th phoneme key vectorAnd a firstlU-th phoneme value vector of layer:
In the formula (1) and the formula (2),is shown aslThe key matrix of the layer(s),is shown aslA matrix of values for the layers; reLU denotes the activation function; "·" denotes the multiplication of a matrix and a vector;
step 3.2.2, saidlThe phoneme memory network of the layer calculates the second by using the formula (4)lU-th phoneme weight of layer:
In the formula (3), "·" represents a vector inner product;
step 3.2.3, saidlThe hierarchical phoneme memory network calculates a weighted average vector using equation (5):
Step 3.2.4, saidlThe phoneme memory network of the layer is obtained by using the formula (6)lLayer jth letter reset vector:
In the formula (5), sigmoid represents an activation function,andrespectively representlA first reset matrix and a second reset matrix of the layer,is shown aslA reset offset vector of a layer;
step 3.2.5, saidlThe phoneme memory network of the layer is obtained by using the formula (7)lLayer enhanced jth letter code vectorSo that the enhanced jth letter code vector is output from the phoneme memory network of L layer|l=1,2, \ 8230;, L }, and thus enhanced n i Letter code vector|l=1,2,…,L, j=1,2,…,n i And is marked as the ith word X i Alpha-code matrix H i ;
4. An electronic device comprising a memory and a processor, wherein the memory is configured to store a program that enables the processor to perform the transliteration method of any of claims 1-3, and wherein the processor is configured to execute the program stored in the memory.
5. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the transliteration method according to any one of claims 1 to 3.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211595293.3A CN115662392B (en) | 2022-12-13 | 2022-12-13 | Transliteration method based on phoneme memory, electronic equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211595293.3A CN115662392B (en) | 2022-12-13 | 2022-12-13 | Transliteration method based on phoneme memory, electronic equipment and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115662392A true CN115662392A (en) | 2023-01-31 |
CN115662392B CN115662392B (en) | 2023-04-25 |
Family
ID=85019419
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211595293.3A Active CN115662392B (en) | 2022-12-13 | 2022-12-13 | Transliteration method based on phoneme memory, electronic equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115662392B (en) |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB1522851A (en) * | 1974-07-17 | 1978-08-31 | Threshold Tech | Apparatus and method for recognizing words from among continuous speech |
CN103020046A (en) * | 2012-12-24 | 2013-04-03 | 哈尔滨工业大学 | Name transliteration method on the basis of classification of name origin |
-
2022
- 2022-12-13 CN CN202211595293.3A patent/CN115662392B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB1522851A (en) * | 1974-07-17 | 1978-08-31 | Threshold Tech | Apparatus and method for recognizing words from among continuous speech |
CN103020046A (en) * | 2012-12-24 | 2013-04-03 | 哈尔滨工业大学 | Name transliteration method on the basis of classification of name origin |
Non-Patent Citations (2)
Title |
---|
宋彦: "一种基于样本和特征选择的语种识别方法" * |
金马等: "基于卷积神经网络的语种识别系统" * |
Also Published As
Publication number | Publication date |
---|---|
CN115662392B (en) | 2023-04-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111897949B (en) | Guided text abstract generation method based on Transformer | |
CN111368565B (en) | Text translation method, text translation device, storage medium and computer equipment | |
CN110059188B (en) | Chinese emotion analysis method based on bidirectional time convolution network | |
CN111460807B (en) | Sequence labeling method, device, computer equipment and storage medium | |
CN111859978A (en) | Emotion text generation method based on deep learning | |
CN108460028B (en) | Domain adaptation method for integrating sentence weight into neural machine translation | |
CN111291534A (en) | Global coding method for automatic summarization of Chinese long text | |
CN112348911B (en) | Semantic constraint-based method and system for generating fine-grained image by stacking texts | |
CN110362797B (en) | Research report generation method and related equipment | |
CN110619127A (en) | Mongolian Chinese machine translation method based on neural network turing machine | |
WO2023226292A1 (en) | Method for extracting relation from text, relation extraction model, and medium | |
CN112561718A (en) | Case microblog evaluation object emotion tendency analysis method based on BilSTM weight sharing | |
CN113128206A (en) | Question generation method based on word importance weighting | |
CN114168754A (en) | Relation extraction method based on syntactic dependency and fusion information | |
CN111401037A (en) | Natural language generation method and device, electronic equipment and storage medium | |
WO2020040255A1 (en) | Word coding device, analysis device, language model learning device, method, and program | |
CN113191150B (en) | Multi-feature fusion Chinese medical text named entity identification method | |
CN109979461A (en) | A kind of voice translation method and device | |
CN114528944B (en) | Medical text coding method, device, equipment and readable storage medium | |
CN115662392A (en) | Transliteration method based on phoneme memory, electronic equipment and storage medium | |
CN115270792A (en) | Medical entity identification method and device | |
CN114580376A (en) | Chinese abstract generating method based on component sentence method analysis | |
CN111428509B (en) | Latin letter-based Uygur language processing method and system | |
CN112818688A (en) | Text processing method, device, equipment and storage medium | |
CN110825869A (en) | Text abstract generating method of variation generation decoder based on copying mechanism |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |