CN106528858A - Lyrics generating method and device - Google Patents
Lyrics generating method and device Download PDFInfo
- Publication number
- CN106528858A CN106528858A CN201611079390.1A CN201611079390A CN106528858A CN 106528858 A CN106528858 A CN 106528858A CN 201611079390 A CN201611079390 A CN 201611079390A CN 106528858 A CN106528858 A CN 106528858A
- Authority
- CN
- China
- Prior art keywords
- lyrics
- generated
- sentence
- lyric
- hidden state
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 60
- 239000013598 vector Substances 0.000 claims abstract description 87
- 230000002457 bidirectional effect Effects 0.000 claims description 33
- 230000002441 reversible effect Effects 0.000 claims description 28
- 230000000306 recurrent effect Effects 0.000 claims description 26
- 238000013528 artificial neural network Methods 0.000 claims description 16
- 238000012549 training Methods 0.000 claims description 12
- 238000006243 chemical reaction Methods 0.000 claims description 7
- 230000006403 short-term memory Effects 0.000 abstract 1
- 238000003062 neural network model Methods 0.000 description 20
- 238000010586 diagram Methods 0.000 description 13
- 230000008569 process Effects 0.000 description 12
- 238000005516 engineering process Methods 0.000 description 9
- 238000004590 computer program Methods 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 241000238631 Hexapoda Species 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 241000254158 Lampyridae Species 0.000 description 1
- 241000220317 Rosa Species 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000007664 blowing Methods 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 235000012736 patent blue V Nutrition 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/60—Information retrieval; Database structures therefor; File system structures therefor of audio data
- G06F16/68—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/683—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/685—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using automatically derived transcript of audio data, e.g. lyrics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/205—Parsing
- G06F40/211—Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Library & Information Science (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Databases & Information Systems (AREA)
- Machine Translation (AREA)
Abstract
The invention provides a lyrics generating method and device. The method comprises the following steps: S1, acquiring source lyrics, and determining the number S of sentences and the length of each sentence in the source lyrics; S2, encoding input sequences consisting of subject terms of a sentence to be generated, the length of the sentence to be generated and generated lyrics based on a long short-term memory (LSTM) model in order to convert the input sequences into a group of hidden states; S3, decoding the hidden states based on the LSTM model including an internal state vector in order to generate lyrics of the sentence to be generated; and S4, repeating the steps S2 and S3 in order to generate S sentences of lyrics. Through adoption of the lyrics generating method and device, the length of each generated sentence of lyrics can be controlled accurately; the subject terms are allocated to each sentence, so that the relevance between the generated lyrics and the subject terms is enhanced; and subsequent sentences are generated by the generated sentences, so that logical relevance among sentences is enhanced.
Description
Technical Field
The invention relates to the technical field of audio processing, in particular to a lyric generating method and device.
Background
It is well known that songs are composed of a combination of lyrics and a piece of music. Lyrics, which are an expression of poetry, are the instinct of songs. With the continuous development of science and technology, the technology of automatically generating lyrics by a machine is gradually mature. At present, the lyric structure of the existing song can be simulated to generate new lyrics, the generated lyrics are ensured to be consistent with the words of the original lyrics, and the generated lyrics can sing according to the music of the original song. Therefore, how to generate lyrics matching a music piece becomes a very challenging technical problem given the music piece.
In the existing lyric generation technology based on a Recurrent Neural Network (RNN) language model, the starting part of the lyric needs to be given first, the starting part of the lyric is used as input, the probability of all characters in a dictionary is output by using the RNN language model, and the character with the maximum probability is used as the next character. The newly generated word is then used as input to continue generating the next word. And repeating the steps until the last character is generated, and finishing the lyric generation process.
However, the existing lyric generation technology based on the language model is difficult to ensure the correlation between the generated lyrics and the theme of the lyrics required by the user, and the generated lyrics have low logic correlation.
Disclosure of Invention
The object of the present invention is to solve at least to some extent one of the above mentioned technical problems.
To this end, a first object of the present invention is to provide a lyric generating method capable of accurately controlling the length of each sentence of lyrics generated, and improving the correlation between the generated lyrics and subject words and the logical correlation between sentences.
A second object of the present invention is to provide a lyric generating apparatus.
A third object of the present invention is to provide a terminal.
A fourth object of the invention is to propose a non-transitory computer-readable storage medium.
A fifth object of the invention is to propose a computer program product.
In order to achieve the above object, an embodiment of a first aspect of the present invention provides a lyric generating method, including: s1, acquiring the source lyrics, and determining the number S of sentences in the source lyrics and the length of each sentence; s2, coding an input sequence consisting of subject words of the sentence to be generated, the length of the sentence to be generated and the generated lyrics based on the two-way time recurrent neural network LSTM model so as to convert the input sequence into a group of hidden states; s3, decoding the hidden state based on the time recursive neural network LSTM model containing the internal state vector to generate the lyrics of the sentence to be generated; and S4, repeating the steps S2 and S3 to generate S sentences of lyrics.
The lyric generating method provided by the embodiment of the first aspect of the invention is characterized in that the number of sentences and the length of each sentence in the source lyrics are determined, an input sequence consisting of subject words and lengths of the sentences to be generated and the generated lyrics is coded and converted into a group of hidden states based on a bidirectional time recurrent neural network model, the hidden states are decoded based on the time recurrent neural network model containing internal state vectors to generate the lyrics of the sentences to be generated, and then the new lyrics with the same number as the source lyric sentences are generated. Therefore, the length of each sentence of generated lyrics can be accurately controlled, the correlation between the generated lyrics and the subject word is improved by distributing the subject word to each sentence, and the subsequent sentence is generated by using the generated sentence, so that the logic correlation between the sentences is improved.
In order to achieve the above object, a second embodiment of the present invention provides a lyric generating apparatus, including: the determining module is used for acquiring the source lyrics and determining the number S of sentences in the source lyrics and the length of each sentence; the conversion module is used for coding a subject word of a sentence to be generated, the length of the sentence to be generated and a generated input sequence of components based on a bidirectional time recursive neural network (LSTM) model so as to convert the input sequence into a group of hidden states; and the generating module is used for decoding the hidden state based on the time recursive neural network LSTM model containing the internal state vector so as to generate the lyrics of the sentence to be generated.
The lyric generating device provided by the embodiment of the second aspect of the invention codes and converts an input sequence consisting of subject words and lengths of sentences to be generated and generated lyrics into a group of hidden states by determining the number of sentences and the length of each sentence in source lyrics based on a bidirectional time recurrent neural network model, decodes the hidden states based on the time recurrent neural network model containing internal state vectors to generate lyrics of the sentences to be generated, and further generates new lyrics with the same number as the source lyric sentences. Therefore, the length of each sentence of generated lyrics can be accurately controlled, the correlation between the generated lyrics and the subject word is improved by distributing the subject word to each sentence, and the subsequent sentence is generated by using the generated sentence, so that the logic correlation between the sentences is improved.
In order to achieve the above object, an embodiment of a third aspect of the present invention provides a terminal, including: a processor; a memory for storing processor-executable instructions. Wherein the processor is configured to perform the steps of:
s1', obtaining the source lyrics, and determining the number S of sentences in the source lyrics and the length of each sentence;
s2', based on the LSTM model of the bidirectional time recursion neural network, encoding an input sequence composed of subject words of the sentence to be generated, the length of the sentence to be generated and the generated lyrics so as to convert the input sequence into a group of hidden states;
s3', decoding the hidden state based on the time recursive neural network LSTM model containing the internal state vector to generate the lyrics of the sentence to be generated;
s4 ', repeat steps S2 ' and S3 ' to generate S sentences of lyrics.
The terminal provided by the embodiment of the third aspect of the invention codes and converts an input sequence consisting of subject words and lengths of sentences to be generated and generated lyrics into a group of hidden states by determining the number of sentences and the length of each sentence in source lyrics based on a bidirectional time recurrent neural network model, decodes the hidden states based on the time recurrent neural network model containing internal state vectors to generate lyrics of the sentences to be generated, and further generates new lyrics with the same number as the sentences of the source lyrics. Therefore, the length of each sentence of generated lyrics can be accurately controlled, the correlation between the generated lyrics and the subject word is improved by distributing the subject word to each sentence, and the subsequent sentence is generated by using the generated sentence, so that the logic correlation between the sentences is improved.
In order to achieve the above object, a fourth embodiment of the present invention provides a non-transitory computer-readable storage medium for storing one or more programs, which when executed by a processor of a mobile terminal, enable the mobile terminal to execute the lyric generating method provided in the first embodiment.
The non-transitory computer readable storage medium according to the fourth aspect of the present invention is configured to determine the number of sentences in the source lyrics and the length of each sentence, encode and convert an input sequence consisting of a subject word and the length of a sentence to be generated and generated lyrics into a set of hidden states based on a bidirectional time recurrent neural network model, decode the hidden states based on a time recurrent neural network model including an internal state vector to generate lyrics of the sentence to be generated, and generate new lyrics having the same number as the sentences of the source lyrics. Therefore, the length of each sentence of generated lyrics can be accurately controlled, the correlation between the generated lyrics and the subject word is improved by distributing the subject word to each sentence, and the subsequent sentence is generated by using the generated sentence, so that the logic correlation between the sentences is improved.
In order to achieve the above object, a fifth embodiment of the present invention provides a computer program product, wherein when instructions in the computer program product are executed by a processor, the method for generating lyrics as proposed in the first embodiment is performed.
The computer program product according to the fifth aspect of the present invention is configured to determine the number of sentences in the source lyrics and the length of each sentence, encode and convert an input sequence composed of subject words and lengths of the sentences to be generated and generated lyrics into a set of hidden states based on a bidirectional time recurrent neural network model, decode the hidden states based on the time recurrent neural network model including internal state vectors to generate lyrics of the sentences to be generated, and generate new lyrics having the same number as the source lyrics sentences. Therefore, the length of each sentence of generated lyrics can be accurately controlled, the correlation between the generated lyrics and the subject word is improved by distributing the subject word to each sentence, and the subsequent sentence is generated by using the generated sentence, so that the logic correlation between the sentences is improved.
Drawings
The foregoing and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
fig. 1 is a schematic flow chart of a lyric generating method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a process of encoding and decoding to generate lyrics of a sentence to be generated based on an LSTM model;
FIG. 3 is a schematic flow diagram of converting an input sequence to a hidden state based on an LSTM model;
FIG. 4 is a schematic flow chart of decoding hidden states based on an LSTM model containing internal state vectors to generate lyrics of a sentence to be generated;
FIG. 5 is a flowchart illustrating a lyric generating method according to another embodiment of the present invention;
FIG. 6 is a flowchart illustrating a lyric generating method according to another embodiment of the present invention;
fig. 7 is a schematic structural diagram of a lyric generating apparatus according to an embodiment of the present invention;
fig. 8 is a schematic structural diagram of a lyric generating apparatus according to another embodiment of the present invention;
fig. 9 is a schematic structural diagram of a lyric generating apparatus according to yet another embodiment of the present invention;
fig. 10 is a schematic structural diagram of a lyric generating apparatus according to a further embodiment of the present invention;
fig. 11 is a schematic structural diagram of a lyric generating apparatus according to still another embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention. On the contrary, the embodiments of the invention include all changes, modifications and equivalents coming within the spirit and terms of the claims appended hereto.
With the continuous development of science and technology, the technology for automatically generating lyrics by a machine is gradually mature. However, the automatic music composing technique is not mature yet, and the composer is still required to compose the lyric music. If the music is not written by the composition family, the generated lyrics cannot become the songs to be sung. In this case, the composition problem can be circumvented by a method of word filling for existing songs. That is, the lyric structure of the existing song can be simulated to generate new lyrics, the generated lyrics are ensured to be consistent with the words of the original lyrics, and the generated lyrics can sing according to the music of the original song. Therefore, given a music piece and a keyword, how to generate lyrics matching the music piece becomes a very challenging technical problem.
The existing technology for generating lyrics based on an RNN language model needs to give the starting part of the lyrics, take the starting part of the lyrics as input, output the probability of all words in a dictionary by using the RNN language model, and take the word with the maximum probability as the next word. The newly generated word is then used as input to continue generating the next word. And repeating the steps until the last character is generated, and finishing the lyric generation process.
However, the existing lyric generation technology based on the language model has difficulty in ensuring the correlation between the lyrics and the keywords due to the need of providing the beginning part of the lyrics. Due to the limited historical information that the language model can remember, it may result in low logical association between sentences. In addition, since the length of the generated content cannot be controlled by generating the lyrics by the prior art, it is difficult to ensure the consistency of the length of the generated lyrics with the length of the specified music.
In order to make up the defects of the existing lyric word generation technology, the invention provides a lyric generation method and a device, which can accurately control the length of each generated lyric, and improve the correlation between the generated lyrics and subject words and the logic correlation between sentences.
Fig. 1 is a schematic flow chart of a lyric generating method according to an embodiment of the present invention.
As shown in fig. 1, the lyric generating method of the present embodiment includes:
s1: the method comprises the steps of obtaining source lyrics, and determining the number S of sentences in the source lyrics and the length of each sentence.
In the embodiment, in order to generate new lyrics, first, source lyrics are acquired, and the number of all sentences in the source lyrics is determined and recorded as S; and simultaneously acquiring the length of each sentence in the source lyrics.
Wherein, the source lyrics refer to the lyrics which are combined with the music to be used into a song; the length of a sentence refers to the number of words contained in the sentence.
In this embodiment, in order to combine the music to be used and the newly generated lyrics into a song for singing and ensure that the newly generated lyrics are matched with the music to be used, it is necessary to first obtain the source lyrics that are combined into the song with the music to be used, and determine the number of sentences in the source lyrics and the length of each sentence.
S2: and coding an input sequence consisting of subject words of the sentence to be generated, the length of the sentence to be generated and the generated lyrics based on a bidirectional time recurrent neural network (LSTM) model so as to convert the input sequence into a group of hidden states.
In this embodiment, after the number of sentences of the source lyrics and the length of each sentence are determined, an input sequence composed of the subject word of the sentence to be generated, the length of the sentence to be generated, and the generated lyrics may be encoded based on a Long-Short Term Memory (LSTM) model, so as to convert the input sequence into a set of hidden states.
Among them, the LSTM model is an improved model of the RNN model. In the LSTM model, conventional neurons in the RNN model are replaced with memory cells, each associated with an input gate, an output gate, and a step of crossing time to feed its internal state undisturbed. Compared with the RNN model, the LSTM model has stronger memory capability and can remember longer historical information.
It should be noted that the subject term is a keyword related to a sentence to be generated, which is pre-assigned according to a user requirement, and may be a single word or a word of a plurality of word groups, which is not limited herein. The length of the sentence to be generated is the same as the length of the sentence at the corresponding position in the source lyric, i.e. the number of words contained in the sentence to be generated is the same as the number of words contained in the sentence at the corresponding position in the source lyric.
In the embodiment, a subject word is firstly allocated to a sentence to be generated, and the corresponding sentence length of the lyric to be generated is determined according to the sentence length of the source lyric. Then, the subject words of the sentence to be generated, the length of the sentence to be generated and the generated lyrics are combined into an input sequence according to the sequence, and the combined input sequence is encoded by utilizing a bidirectional LSTM model so as to convert the input sequence into a group of hidden states.
For example, suppose that after the subject word of the sentence to be generated, the length of the sentence to be generated and the generated lyrics are combined in sequence, an input sequence with a length of T is formed, and is marked as X (X ═ X1，x2，…，xT). Inputting the sequence into a bidirectional LSTM model for coding to obtain a group of hidden states marked as (h)1，h2，…，hT)。
It should be noted that, when the lyrics of the first sentence to be generated are generated, since there is no generated lyric, only the input sequence composed of the subject word of the sentence to be generated and the length of the sentence to be generated is input to the LSTM model for encoding, so as to convert the input sequence into a set of hidden states.
S3: the hidden state is decoded based on a temporal recurrent neural network LSTM model containing internal state vectors to generate the lyrics of the sentence to be generated.
In this embodiment, after obtaining the hidden state corresponding to the input sequence composed of the subject word and the length of the sentence to be generated and the generated lyrics, the obtained hidden state is decoded based on the LSTM model including the internal state vector, and the lyrics of the sentence to be generated can be generated.
The internal state vector is a parameter carried by the LSTM model at the decoding end, and is used for generating a word most relevant to the input hidden state, and the word is continuously updated in the decoding process, and the specific updating process is given in the subsequent content.
As an example, referring to fig. 2, fig. 2 is a schematic diagram of a process of generating lyrics of a sentence to be generated by performing encoding and decoding based on the LSTM model.
As shown in fig. 2, the subject word of the sentence to be generated is "season",the length of a sentence to be generated is '7', the generated lyrics are 'I turn north all the way', an input sequence is formed in sequence by 'season', '7', 'I', 'one', 'way', 'north' and 'north', the length of the input sequence is 7, and is marked as X ═ (X ═ X1，x2，…，x7). Then, the input sequence with the length of 7 is coded based on the bidirectional LSTM model to obtain a group of hidden states, which are marked as (h)1，h2，…，h7). Then, the hidden states are decoded based on an LSTM model including an internal state vector according to the attention scores corresponding to the hidden states when the ith word is generated, and the ith word is generated. Wherein i is a positive integer, i is 1, 2, …, 7. In FIG. 2, a7,1，a7,2，…，a7,6，a7,7Representing the attention scores corresponding to the hidden states when the 7 th word is generated, after weighted summation, combining the internal state vectors S7Decoding based on LSTM model to obtain the 7 th word y of the sentence to be generated7I.e., a "section" word.
It should be noted that the attention score corresponding to the hidden state is obtained by calculation, and a specific calculation process will be given in the following.
S4: steps S2 and S3 are repeated to generate S sentence lyrics.
In this embodiment, after the lyrics of the current sentence to be generated are generated, the lyrics to be generated later are used as a new sentence to be generated, the subject word and the length of the sentence to be generated are determined, the newly generated lyrics are used as the generated lyrics, the foregoing steps S2 and S3 are repeated until the lyrics of the last sentence to be generated are generated, and finally, the lyrics of S sentences are obtained.
Take song "Chong Er Fei" as an example. In order to generate new lyrics of a piece of music that can use the song "fly for worm", it is necessary to ensure that the newly generated lyrics have the same lyric structure as the lyrics of the song "fly for worm". The specific process of generating new lyrics is described as follows:
first, the lyrics of the song "worm flight" need to be obtained as the source lyrics. The lyrics of the song "Chonger Fei" are as follows:
dark sky sag
Bright stars following each other
Flying insect and flying baby fly
Who you are thinking
Tearing by stars on the sky
Overground rose withering
Cold air blowing
As long as you accompany
By acquiring the source lyrics of the insect fly, the number of sentences of the source lyrics can be determined to be 8, and the length of each sentence is 7, 7, 6, 5, 7, 7, 6, 5.
Assume that the subject word of the first sentence to be generated is "sky". According to the length of the first sentence of the source lyrics, the length of the sentence to be generated in the first sentence can be determined to be 7. Because the sentence to be generated is the first sentence and no lyrics are generated before, the sky and the 7 form an input sequence, and the first sentence of lyrics, namely the sky with blue hugs, of the new lyrics can be obtained after the two-way LSTM model is used for coding and the LSTM model containing the internal state vector is used for decoding. Suppose the subject word of the sentence to be generated in the second sentence is "stars". According to the length of the second sentence of the source lyrics, the length of the sentence to be generated in the second sentence can be determined to be 7. The subject words and the length of the sentence to be generated in the second sentence and the generated lyrics, namely the sky with blue embrace form an input sequence, and after the input sequence is coded by a bidirectional LSTM model and decoded by an LSTM model containing an internal state vector, the second sentence of lyrics of the new lyrics can be obtained, namely the lyrics have twinkling with a star. Repeating the steps, 8 words of new words with the same structure as the source words can be obtained, as follows:
sky blue
With a star flashing
Firefly
Free from the world
Over mountains and oceans
Dream leaning on your shoulder
Wind wave and wind wave
Get lost of direction
It should be noted that, since the method for generating lyrics proposed in this embodiment uses the LSTM model, each subsequent lyric is generated based on all the previous lyrics. That is, when the fifth sentence of lyrics is generated, the first four sentences of lyrics that have already been generated are used; when the sixth lyric is generated, the first five lyrics which are already generated are used. In this way, the relevance between sentences can be made stronger.
The lyric generating method provided by the embodiment of the invention is characterized in that the number of sentences in the source lyrics and the length of each sentence are determined, an input sequence consisting of subject words and lengths of the sentences to be generated and the generated lyrics is coded and converted into a group of hidden states based on a bidirectional time recurrent neural network model, the hidden states are decoded based on the time recurrent neural network model containing internal state vectors to generate the lyrics of the sentences to be generated, and further, the number of the lyrics is the same as that of the sentences of the source lyrics. Therefore, the length of each sentence of generated lyrics can be accurately controlled, the correlation between the generated lyrics and the subject word is improved by distributing the subject word to each sentence, and the subsequent sentence is generated by using the generated sentence, so that the logic correlation between the sentences is improved.
FIG. 3 is a flow diagram of converting an input sequence to a hidden state based on the LSTM model.
As shown in fig. 3, the step S2 may specifically include the following steps:
s21: and forward coding the input sequence based on a forward LSTM model to obtain a forward hidden state.
S22: the input sequence is reverse encoded based on a reverse LSTM model to obtain a reverse hidden state.
In the embodiment, after an input sequence consisting of subject words of a sentence to be generated, the length of the sentence to be generated and generated lyrics is input into a bidirectional LSTM model, forward coding is carried out on the input sequence based on the forward LSTM model to generate a forward hidden state; the input sequence is reverse encoded based on a reverse LSTM model to generate a reverse hidden state.
It should be noted that, in this embodiment, the steps of obtaining the forward hidden state and the reverse hidden state may be performed sequentially or simultaneously, which is not limited in the present invention.
S23: and splicing the forward hidden state and the reverse hidden state to generate the hidden state.
In this embodiment, after the forward hidden state and the reverse hidden state are obtained respectively, the obtained forward hidden state and the obtained reverse hidden state are spliced to generate a group of hidden states corresponding to the input sequence.
According to the lyric generation method provided by the embodiment of the invention, the forward hidden state is obtained by forward coding the input sequence based on the forward LSTM model, the reverse hidden state is obtained by reverse coding the input sequence based on the reverse LSTM model, and the hidden state is generated by splicing the forward hidden state and the reverse hidden state, so that the information of the upper part and the lower part can be obtained at the same time.
FIG. 4 is a schematic flow chart of decoding hidden states based on an LSTM model containing internal state vectors to generate lyrics of a sentence to be generated.
As shown in fig. 4, the aforementioned step S3 may include the steps of:
s31: and acquiring the attention score corresponding to the hidden state.
In this embodiment, after obtaining the hidden states corresponding to the input sequence, the attention scores corresponding to the hidden states are continuously obtained.
Specifically, the attention score corresponding to the hidden state may be obtained by calculation according to formula (1) and formula (2).
Wherein v isa、WaAnd UaIs a parameter matrix generated and updated during the model training process. k is a positive integer, and k is 1, 2, …, Tx，TxIndicating the length of the input sequence X. i. j is a positive integer, wherein i is 1, 2, …, TxRepresenting the ith element of the input sequence X; j-1, 2, …, TxThe j-th element of the hidden state corresponding to the input sequence X is represented.
S32: and obtaining a history information vector of the input sequence according to the hidden state and the attention score corresponding to the hidden state.
In this embodiment, after obtaining the attention score corresponding to the hidden state, the history information vector of the input sequence may be obtained according to the attention scores corresponding to the hidden state and the hidden state.
Specifically, the history information vector of the input sequence may be obtained by calculation according to formula (3).
Wherein, i and j are positive integers, i is 1, 2, …, Tx，j＝1，2，…，Tx。
Slave maleAs can be seen from equation (3), when generating the history information vector corresponding to the ith element in the input sequence X, the attention scores corresponding to the hidden states at this time are multiplied by the corresponding hidden states respectively and then added, so as to obtain the history information vector c corresponding to the ith element in the input sequence Xi。
S33: a current word is generated based on the internal state vector, the historical information vector, and a previous word.
In this embodiment, after the history information vector of the input sequence is obtained through calculation, the current word may be generated according to the internal state vector, the history information vector, and the previous word.
Specifically, the current word may be computationally generated according to equation (4).
yi＝arg max p(y|si,ci,yi-1) (4)
Wherein s isiRepresenting an internal state vector; c. CiRepresenting the calculated historical information vector; y isi-1Representing the previous word, i being a positive integer, i ═ 1, 2, …, the length of the sentence to be generated.
S34: and combining the generated current words into lyrics of a sentence to be generated.
In this embodiment, after generating a plurality of words that conform to the length of the sentence to be generated, the generated plurality of words may be combined into the lyrics of the sentence to be generated.
For example, assuming that the length of the sentence to be generated is 7, and the 7 words generated by using the lyric generating method proposed in this embodiment are "away", "having", "you", "of", "season", and "section", respectively, the lyric of the sentence to be generated is "away from your season" by combining the 7 words.
Optionally, referring to fig. 5, fig. 5 is a schematic flowchart of a lyric generating method according to another embodiment of the present invention.
As shown in fig. 5, after generating the current word from the internal state vector, the history information vector and the previous word, the following steps may be further included:
s35: the internal state vector is updated.
In this embodiment, after the current word is generated according to the internal state vector, the history information vector and the previous word, the internal state vector may be updated to improve the accuracy of generating the word.
Specifically, the internal state vector of the decoding-side LSTM model may be updated according to equation (5).
si＝f(si-1,ci-1,yi) (5)
Where i is a positive integer, i is 1, 2, …, the length of the sentence to be generated.
It should be noted that, in this embodiment, step S34 and step S35 may be performed simultaneously or sequentially, and the present invention is not limited to this.
According to the lyric generating method provided by the embodiment of the invention, the attention score corresponding to the hidden state is obtained, the historical information vector of the input sequence is obtained according to the attention score and the corresponding hidden state, the current word is generated according to the internal state vector, the historical information vector and the previous word, and the generated words are combined into the lyric of the sentence to be generated, so that the relevance between the words can be improved, and the accuracy of lyric generation is further improved.
Fig. 6 is a flowchart illustrating a lyric generating method according to another embodiment of the present invention.
As shown in fig. 6, based on the foregoing embodiment, the lyric generating method may further include the following steps:
s5: a bi-directional LSTM model and an LSTM model containing internal state vectors are trained.
In the embodiment of the present invention, in order to generate lyrics by using the bidirectional LSTM model and the LSTM model including the internal state vector, the two models need to be trained.
Specifically, training a bidirectional LSTM model and an LSTM model containing internal state vectors includes: obtaining a subject word sample and a lyric sample corresponding to the subject word sample; the bi-directional LSTM model and the LSTM model containing the internal state vectors are trained from the subject word samples and the lyric samples.
In this embodiment, in order to train the bidirectional LSTM model and the LSTM model including the internal state vector, a large number of lyrics need to be obtained to generate training samples.
Specifically, a subject word is respectively extracted from each lyric in the obtained lyrics, and the extracted subject word, the length of the lyric corresponding to the subject word and the previous lyric are used as a subject word sample; and taking the lyrics of the sentence corresponding to the subject words as lyric samples. The subject word samples and corresponding lyric samples constitute sample pairs.
It should be noted that the lyrics may be obtained from a preset lyric library, may also be obtained from an internet database, or may be obtained in other ways, which is not limited in the present invention.
In addition, it should be noted that the sample pairs of the subject word samples and the lyric samples may be obtained from the lyrics in a manual labeling manner, or the sample pairs of the subject word samples and the lyric samples may be automatically obtained by using a correlation technique, which is not limited in this disclosure.
The construction process of the sample pair is described below by taking the lyrics of the first 4 sentences of the song "one way to north" as an example.
First, a word is extracted from each sentence of lyrics as a subject word, as shown in table 1:
TABLE 1 corresponding relationship table of subject word and lyric
| Subject term | Length of lyrics | Lyrics corresponding to subject words |
| Road surface | 5 | I turn all the way to the north |
| Season | 7 | Away from the season of your |
| Tired of | 5 | You say you are tired |
| Love | 7 | Who can no longer love |
Next, sample pairs of subject word samples and lyric samples were constructed as shown in table 2:
table 2 sample pair table
| Subject word sample | Lyric sample |
| Road 5 | I turn all the way to the north |
| Season 7, i all the way to the north | Away from the season of your |
| Accumulate 5 my season of leaving your one way to the north | You say you are tired |
| Love 7I leave your season all the way to north you say you are so tired | Who can no longer love |
As can be seen from table 2, the subject word sample consists of the subject word, the length of the lyric corresponding to the subject word and the lyric of the previous sentence, and the lyric sample consists of the lyric corresponding to the subject word. And since there is no lyric before the first sentence of lyrics, the subject word sample corresponding to the first sentence of lyrics consists only of the subject word of the first sentence of lyrics and the length of the first sentence of lyrics.
After a large number of sample pairs consisting of the subject word samples and the lyric samples corresponding to the subject word samples are obtained, the bidirectional LSTM model and the LSTM model containing the internal state vector can be trained according to the obtained sample pairs.
It should be noted that, the step S5 in the present embodiment may be executed at any time before the step S2 is executed, which is not limited by the present invention.
According to the lyric generating method provided by the embodiment of the invention, the bidirectional LSTM model and the LSTM model containing the internal state vector are trained by obtaining the subject word sample and the corresponding lyric sample, so that the bidirectional LSTM model and the LSTM model containing the internal state vector can be used for automatically generating lyrics, and the lyric generating accuracy is further improved.
In order to implement the foregoing embodiment, the present invention further provides a lyric generating apparatus, and fig. 7 is a schematic structural diagram of the lyric generating apparatus according to an embodiment of the present invention.
As shown in fig. 7, the lyric generating apparatus of the present embodiment includes: a determination module 710, a conversion module 720, and a generation module 730. Wherein,
the determining module 710 is configured to obtain the source lyrics, and determine the number S of sentences in the source lyrics and the length of each sentence.
And the conversion module 720 is configured to encode an input sequence composed of a subject word of the sentence to be generated, the length of the sentence to be generated, and the generated lyrics based on the bidirectional time recurrent neural network LSTM model, so as to convert the input sequence into a set of hidden states.
The generating module 730 is configured to decode the hidden state based on the LSTM model including the internal state vector to generate lyrics of the sentence to be generated.
It should be noted that, in order to generate the lyrics of the whole song, the conversion module 720 and the generation module 730 need to repeat the work until S sentences of lyrics are generated.
It should be noted that the explanation of the embodiment of the lyric generating method in the foregoing embodiment is also applicable to the lyric generating device in this embodiment, and the implementation principle is similar, and is not described herein again.
The lyric generating device provided by the embodiment of the invention codes and converts the subject words and the length of the sentences to be generated and the input sequence consisting of the generated lyrics into a group of hidden states by determining the number of the sentences and the length of each sentence in the source lyrics based on the bidirectional time recurrent neural network model, decodes the hidden states based on the time recurrent neural network model containing the internal state vector to generate the lyrics of the sentences to be generated, and further generates new lyrics with the same number as the sentences of the source lyrics. Therefore, the length of each sentence of generated lyrics can be accurately controlled, the correlation between the generated lyrics and the subject word is improved by distributing the subject word to each sentence, and the subsequent sentence is generated by using the generated sentence, so that the logic correlation between the sentences is improved.
Fig. 8 is a schematic structural diagram of a lyric generating apparatus according to another embodiment of the present invention.
As shown in fig. 8, the conversion module 720 of the lyric generating apparatus may include:
the forward coding unit 721 is configured to forward code the input sequence based on a forward LSTM model to obtain a forward hidden state.
An inverse coding unit 722, configured to perform inverse coding on the input sequence based on the inverse LSTM model to obtain an inverse hidden state.
And the splicing unit 723 is used for splicing the forward hidden state and the reverse hidden state to generate a hidden state.
It should be noted that the explanation of the embodiment of the lyric generating method in the foregoing embodiment is also applicable to the lyric generating device in this embodiment, and the implementation principle is similar, and is not described herein again.
The lyric generating device provided by the embodiment of the invention can obtain the information of the upper and the lower text at the same time by carrying out forward coding on the input sequence based on the forward LSTM model to obtain the forward hidden state, carrying out reverse coding on the input sequence based on the reverse LSTM model to obtain the reverse hidden state, and splicing the forward hidden state and the reverse hidden state to generate the hidden state, thereby obtaining more information compared with a one-way LSTM model and an RNN model and further improving the accuracy of lyric generation.
Fig. 9 is a schematic structural diagram of a lyric generating apparatus according to still another embodiment of the present invention.
As shown in fig. 9, the generating module 730 of the lyric generating apparatus may include:
an obtaining unit 731, configured to obtain an attention score corresponding to the hidden state.
An obtaining unit 732, configured to obtain a history information vector of the input sequence according to the hidden state and the attention score corresponding to the hidden state.
A generating unit 733 for generating a current word from the internal state vector, the history information vector and the previous word.
A combining unit 734, configured to combine the generated current word into lyrics of the sentence to be generated.
Optionally, as shown in fig. 10, the generating module 730 of the lyric generating apparatus may further include:
an updating unit 735 is configured to update the internal state vector after generating the current word from the internal state vector, the history information vector and the previous word.
It should be noted that the explanation of the embodiment of the lyric generating method in the foregoing embodiment is also applicable to the lyric generating device in this embodiment, and the implementation principle is similar, and is not described herein again.
According to the lyric generating device provided by the embodiment of the invention, the attention score corresponding to the hidden state is obtained, the historical information vector of the input sequence is obtained according to the attention score and the corresponding hidden state, the current word is generated according to the internal state vector, the historical information vector and the previous word, and the generated words are combined into the lyric of the sentence to be generated, so that the relevance between the words can be improved, and the accuracy of lyric generation is further improved.
Fig. 11 is a schematic structural diagram of a lyric generating apparatus according to still another embodiment of the present invention.
As shown in fig. 11, the lyric generating apparatus may further include:
a training module 740 for training the bi-directional LSTM model and the LSTM model containing the internal state vectors.
Specifically, training module 740 is configured to:
obtaining a subject word sample and a lyric sample corresponding to the subject word sample;
the bi-directional LSTM model and the LSTM model containing the internal state vectors are trained from the subject word samples and the lyric samples.
It should be noted that the explanation of the embodiment of the lyric generating method in the foregoing embodiment is also applicable to the lyric generating device in this embodiment, and the implementation principle is similar, and is not described herein again.
The lyric generating device provided by the embodiment of the invention trains the bidirectional LSTM model and the LSTM model containing the internal state vector by obtaining the subject word sample and the corresponding lyric sample, can automatically generate lyrics by utilizing the bidirectional LSTM model and the LSTM model containing the internal state vector, and further improves the accuracy of lyric generation.
In order to implement the above embodiments, the present invention further provides a terminal, including: a processor, and a memory for storing processor-executable instructions. Wherein the processor is configured to perform the steps of:
s1', obtaining the source lyrics, and determining the number S of sentences in the source lyrics and the length of each sentence;
s2', based on the LSTM model of the bidirectional time recursion neural network, encoding an input sequence composed of subject words of the sentence to be generated, the length of the sentence to be generated and the generated lyrics so as to convert the input sequence into a group of hidden states;
s3', decoding the hidden state based on the time recursive neural network LSTM model containing the internal state vector to generate the lyrics of the sentence to be generated;
s4 ', repeat steps S2 ' and S3 ' to generate S sentences of lyrics.
It should be noted that the explanation of the embodiment of the method for generating a lyric in the foregoing embodiment is also applicable to the terminal in this embodiment, and the implementation principle is similar, and is not described herein again.
The terminal provided by the embodiment of the invention codes and converts the subject words and the length of the sentences to be generated and the input sequence consisting of the generated lyrics into a group of hidden states by determining the number of the sentences in the source lyrics and the length of each sentence based on the bidirectional time recurrent neural network model, decodes the hidden states based on the time recurrent neural network model containing internal state vectors to generate the lyrics of the sentences to be generated, and further generates new lyrics with the same number as the sentences of the source lyrics. Therefore, the length of each sentence of generated lyrics can be accurately controlled, the correlation between the generated lyrics and the subject word is improved by distributing the subject word to each sentence, and the subsequent sentence is generated by using the generated sentence, so that the logic correlation between the sentences is improved.
In order to implement the foregoing embodiments, the present invention further proposes a non-transitory computer-readable storage medium storing one or more programs, which when executed by a processor of a mobile terminal, enable the mobile terminal to execute the lyric generating method proposed by the first aspect of the present invention.
The non-transitory computer readable storage medium provided by the embodiment of the invention is used for coding and converting an input sequence consisting of subject words and lengths of sentences to be generated and generated lyrics into a group of hidden states by determining the number of sentences and the length of each sentence in source lyrics based on a bidirectional time recurrent neural network model, decoding the hidden states based on the time recurrent neural network model containing internal state vectors to generate lyrics of the sentences to be generated, and further generating new lyrics with the same number as the sentences of the source lyrics. Therefore, the length of each sentence of generated lyrics can be accurately controlled, the correlation between the generated lyrics and the subject word is improved by distributing the subject word to each sentence, and the subsequent sentence is generated by using the generated sentence, so that the logic correlation between the sentences is improved.
In order to implement the foregoing embodiments, the present invention further provides a computer program product, and when instructions in the computer program product are executed by a processor, the computer program product executes the lyric generating method according to the first aspect of the present invention.
The computer program product provided by the embodiment of the invention codes and converts an input sequence consisting of subject words and lengths of sentences to be generated and generated lyrics into a group of hidden states by determining the number of sentences and the length of each sentence in source lyrics based on a bidirectional time recurrent neural network model, decodes the hidden states based on the time recurrent neural network model containing internal state vectors to generate the lyrics of the sentences to be generated, and further generates new lyrics with the same number as the sentences of the source lyrics. Therefore, the length of each sentence of generated lyrics can be accurately controlled, the correlation between the generated lyrics and the subject word is improved by distributing the subject word to each sentence, and the subsequent sentence is generated by using the generated sentence, so that the logic correlation between the sentences is improved.
It should be noted that the terms "first," "second," and the like in the description of the present invention are used for descriptive purposes only and are not to be construed as indicating or implying relative importance. In addition, in the description of the present invention, "a plurality" means two or more unless otherwise specified.
Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps of the process, and alternate implementations are included within the scope of the preferred embodiment of the present invention in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the present invention.
It should be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.
It will be understood by those skilled in the art that all or part of the steps carried by the method for implementing the above embodiments may be implemented by hardware related to instructions of a program, which may be stored in a computer readable storage medium, and when the program is executed, the program includes one or a combination of the steps of the method embodiments.
In addition, functional units in the embodiments of the present invention may be integrated into one processing module, or each unit may exist alone physically, or two or more units are integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. The integrated module, if implemented in the form of a software functional module and sold or used as a stand-alone product, may also be stored in a computer readable storage medium.
The storage medium mentioned above may be a read-only memory, a magnetic or optical disk, etc.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.
Claims (12)
1. A method for generating lyrics, comprising:
s1, obtaining source lyrics, and determining the number S of sentences in the source lyrics and the length of each sentence;
s2, coding an input sequence consisting of subject words of a sentence to be generated, the length of the sentence to be generated and generated lyrics based on a bidirectional time recursive neural network (LSTM) model so as to convert the input sequence into a group of hidden states;
s3, decoding the hidden state based on a time recursive neural network (LSTM) model containing an internal state vector to generate lyrics of the sentence to be generated;
and S4, repeating the steps S2 and S3 to generate S sentences of lyrics.
2. The method according to claim 1, wherein the step S2 includes:
forward coding the input sequence based on a forward LSTM model to obtain a forward hidden state;
reversely encoding the input sequence based on a reverse LSTM model to obtain a reverse hidden state;
concatenating the forward hidden state and the reverse hidden state to generate the hidden state.
3. The method according to claim 1, wherein the step S3 includes:
acquiring an attention score corresponding to the hidden state;
obtaining a historical information vector of the input sequence according to the hidden state and the attention score corresponding to the hidden state;
generating a current word according to the internal state vector, the historical information vector and a previous word;
and combining the generated current words into the lyrics of the sentence to be generated.
4. The method of claim 3, after generating a current word from the internal state vector, the history information vector, and a previous word, further comprising:
updating the internal state vector.
5. The method of claim 1, further comprising:
the bi-directional LSTM model and the LSTM model containing the internal state vectors are trained.
6. The method of claim 5, wherein training the bi-directional LSTM model and the LSTM model containing internal state vectors comprises:
obtaining a subject word sample and a lyric sample corresponding to the subject word sample;
training the bidirectional LSTM model and the LSTM model containing the internal state vector according to the subject word sample and the lyric sample.
7. A lyric generating apparatus, characterized by comprising:
the system comprises a determining module, a judging module and a judging module, wherein the determining module is used for acquiring source lyrics and determining the number S of sentences in the source lyrics and the length of each sentence;
the conversion module is used for coding an input sequence consisting of subject words of a sentence to be generated, the length of the sentence to be generated and generated lyrics based on a bidirectional time recurrent neural network (LSTM) model so as to convert the input sequence into a group of hidden states;
and the generating module is used for decoding the hidden state based on a time recursive neural network (LSTM) model containing an internal state vector so as to generate the lyrics of the sentence to be generated.
8. The apparatus of claim 7, wherein the conversion module comprises:
the forward coding unit is used for carrying out forward coding on the input sequence based on a forward LSTM model so as to obtain a forward hidden state;
the reverse coding unit is used for performing reverse coding on the input sequence based on a reverse LSTM model so as to obtain a reverse hidden state;
and the splicing unit is used for splicing the forward hidden state and the reverse hidden state to generate the hidden state.
9. The apparatus of claim 7, wherein the generating module comprises:
an obtaining unit, configured to obtain an attention score corresponding to the hidden state;
an obtaining unit, configured to obtain a history information vector of the input sequence according to the hidden state and an attention score corresponding to the hidden state;
a generating unit, configured to generate a current word according to the internal state vector, the history information vector, and a previous word;
and the combination unit is used for combining the generated current words into the lyrics of the sentence to be generated.
10. The apparatus of claim 9, further comprising:
an updating unit for updating the internal state vector after generating a current word from the internal state vector, the history information vector and a previous word.
11. The apparatus of claim 7, further comprising:
and the training module is used for training the bidirectional LSTM model and the LSTM model containing the internal state vector.
12. The apparatus of claim 11, wherein the training module is to:
obtaining a subject word sample and a lyric sample corresponding to the subject word sample;
training the bidirectional LSTM model and the LSTM model containing the internal state vector according to the subject word sample and the lyric sample.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611079390.1A CN106528858A (en) | 2016-11-29 | 2016-11-29 | Lyrics generating method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611079390.1A CN106528858A (en) | 2016-11-29 | 2016-11-29 | Lyrics generating method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106528858A true CN106528858A (en) | 2017-03-22 |
Family
ID=58355275
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611079390.1A Pending CN106528858A (en) | 2016-11-29 | 2016-11-29 | Lyrics generating method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106528858A (en) |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107122492A (en) * | 2017-05-19 | 2017-09-01 | 北京金山安全软件有限公司 | Lyric generation method and device based on picture content |
| CN107168957A (en) * | 2017-06-12 | 2017-09-15 | 云南大学 | A kind of Chinese word cutting method |
| CN107491541A (en) * | 2017-08-24 | 2017-12-19 | 北京丁牛科技有限公司 | File classification method and device |
| CN108153913A (en) * | 2018-01-24 | 2018-06-12 | 北京神州泰岳软件股份有限公司 | Return information generates training method, return information generation method and the device of model |
| CN108304436A (en) * | 2017-09-12 | 2018-07-20 | 深圳市腾讯计算机系统有限公司 | The generation method of style sentence, the training method of model, device and equipment |
| CN108415893A (en) * | 2018-03-15 | 2018-08-17 | 平安科技（深圳）有限公司 | Poem automatic generation method, device, computer equipment and storage medium |
| CN109408630A (en) * | 2018-10-17 | 2019-03-01 | 杭州世平信息科技有限公司 | A method of law court's viewpoint is automatically generated according to the description of crime fact |
| CN109460812A (en) * | 2017-09-06 | 2019-03-12 | 富士通株式会社 | Average information analytical equipment, the optimization device, feature visualization device of neural network |
| CN109582956A (en) * | 2018-11-15 | 2019-04-05 | 中国人民解放军国防科技大学 | text representation method and device applied to sentence embedding |
| CN109740145A (en) * | 2018-11-30 | 2019-05-10 | 平安科技（深圳）有限公司 | Lyrics intelligent generation method, device, storage medium and computer equipment |
| CN109815493A (en) * | 2019-01-09 | 2019-05-28 | 厦门大学 | A kind of modeling method that the intelligence hip-hop music lyrics generate |
| CN110164412A (en) * | 2019-04-26 | 2019-08-23 | 吉林大学珠海学院 | A kind of music automatic synthesis method and system based on LSTM |
| CN110287489A (en) * | 2019-06-24 | 2019-09-27 | 北京大米科技有限公司 | Document creation method, device, storage medium and electronic equipment |
| CN110309275A (en) * | 2018-03-15 | 2019-10-08 | 北京京东尚科信息技术有限公司 | A kind of method and apparatus that dialogue generates |
| CN110362696A (en) * | 2019-06-11 | 2019-10-22 | 平安科技（深圳）有限公司 | Lyrics generation method, system, computer equipment and computer readable storage medium |
| CN110399522A (en) * | 2019-07-03 | 2019-11-01 | 中国传媒大学 | A kind of music singing search method and device based on LSTM and layering and matching |
| CN110555197A (en) * | 2018-05-31 | 2019-12-10 | 清华大学 | Automatic Chinese poem generating method and device |
| CN110688857A (en) * | 2019-10-08 | 2020-01-14 | 北京金山数字娱乐科技有限公司 | Article generation method and device |
| CN111191439A (en) * | 2019-12-16 | 2020-05-22 | 浙江大搜车软件技术有限公司 | Natural sentence generation method and device, computer equipment and storage medium |
| CN111506767A (en) * | 2020-03-04 | 2020-08-07 | 北京灵动音科技有限公司 | Song word filling processing method and device, electronic equipment and storage medium |
| CN111767694A (en) * | 2019-03-26 | 2020-10-13 | 北京京东尚科信息技术有限公司 | Text generation method and device and computer readable storage medium |
| CN111797196A (en) * | 2020-06-01 | 2020-10-20 | 武汉大学 | Service discovery method combining attention mechanism LSTM and neural topic model |
| CN112084310A (en) * | 2019-06-12 | 2020-12-15 | 阿里巴巴集团控股有限公司 | Reply information generation and automatic reply method and device |
| CN112163414A (en) * | 2020-08-27 | 2021-01-01 | 同济大学 | Chinese lyric generating method based on Word2Vec, LSTM and attention mechanism |
| CN112669815A (en) * | 2020-12-17 | 2021-04-16 | 广州华多网络科技有限公司 | Song customization generation method and corresponding device, equipment and medium |
| CN112783334A (en) * | 2019-11-08 | 2021-05-11 | 阿里巴巴集团控股有限公司 | Text generation method and device, electronic equipment and computer-readable storage medium |
| CN113553863A (en) * | 2021-06-30 | 2021-10-26 | 北京百度网讯科技有限公司 | Text generation method and device, electronic equipment and storage medium |
| US11170176B2 (en) | 2019-09-27 | 2021-11-09 | International Business Machines Corporation | Artificial intelligence based word generation |
| KR102368418B1 (en) * | 2020-12-31 | 2022-03-02 | (주)휴에버그린팜 | Apparatus and Method for Creating Rap Lyrics included Punch Lines |
| CN114186545A (en) * | 2021-11-10 | 2022-03-15 | 北京达佳互联信息技术有限公司 | Lyric generation method and device, electronic equipment and storage medium |
| CN115114474A (en) * | 2021-03-17 | 2022-09-27 | 北京达佳互联信息技术有限公司 | Lyric generation model training method, lyric generation method, device and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101253496A (en) * | 2005-07-01 | 2008-08-27 | 微软公司 | Generating Chinese language couplets |
| CN103336803A (en) * | 2013-06-21 | 2013-10-02 | 杭州师范大学 | Method for generating name-embedded spring festival scrolls through computer |
| CN104731775A (en) * | 2015-02-26 | 2015-06-24 | 北京捷通华声语音技术有限公司 | Method and device for converting spoken languages to written languages |
| CN105955953A (en) * | 2016-05-03 | 2016-09-21 | 成都数联铭品科技有限公司 | Word segmentation system |
| CN105955964A (en) * | 2016-06-13 | 2016-09-21 | 北京百度网讯科技有限公司 | Method and apparatus for automatically generating poem |
| CN106095752A (en) * | 2016-06-07 | 2016-11-09 | 北京百度网讯科技有限公司 | A kind of method and apparatus for automatically generating poem |
-
2016
- 2016-11-29 CN CN201611079390.1A patent/CN106528858A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101253496A (en) * | 2005-07-01 | 2008-08-27 | 微软公司 | Generating Chinese language couplets |
| CN103336803A (en) * | 2013-06-21 | 2013-10-02 | 杭州师范大学 | Method for generating name-embedded spring festival scrolls through computer |
| CN104731775A (en) * | 2015-02-26 | 2015-06-24 | 北京捷通华声语音技术有限公司 | Method and device for converting spoken languages to written languages |
| CN105955953A (en) * | 2016-05-03 | 2016-09-21 | 成都数联铭品科技有限公司 | Word segmentation system |
| CN106095752A (en) * | 2016-06-07 | 2016-11-09 | 北京百度网讯科技有限公司 | A kind of method and apparatus for automatically generating poem |
| CN105955964A (en) * | 2016-06-13 | 2016-09-21 | 北京百度网讯科技有限公司 | Method and apparatus for automatically generating poem |
Cited By (47)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107122492A (en) * | 2017-05-19 | 2017-09-01 | 北京金山安全软件有限公司 | Lyric generation method and device based on picture content |
| CN107168957A (en) * | 2017-06-12 | 2017-09-15 | 云南大学 | A kind of Chinese word cutting method |
| CN107491541A (en) * | 2017-08-24 | 2017-12-19 | 北京丁牛科技有限公司 | File classification method and device |
| CN109460812A (en) * | 2017-09-06 | 2019-03-12 | 富士通株式会社 | Average information analytical equipment, the optimization device, feature visualization device of neural network |
| CN108304436A (en) * | 2017-09-12 | 2018-07-20 | 深圳市腾讯计算机系统有限公司 | The generation method of style sentence, the training method of model, device and equipment |
| CN108304436B (en) * | 2017-09-12 | 2019-11-05 | 深圳市腾讯计算机系统有限公司 | Generation method, the training method of model, device and the equipment of style sentence |
| WO2019052311A1 (en) * | 2017-09-12 | 2019-03-21 | 腾讯科技（深圳）有限公司 | Style statement generation method, model training method and apparatus, and computer device |
| US11348570B2 (en) | 2017-09-12 | 2022-05-31 | Tencent Technology (Shenzhen) Company Limited | Method for generating style statement, method and apparatus for training model, and computer device |
| US11869485B2 (en) | 2017-09-12 | 2024-01-09 | Tencent Technology (Shenzhen) Company Limited | Method for generating style statement, method and apparatus for training model, and computer device |
| CN108153913A (en) * | 2018-01-24 | 2018-06-12 | 北京神州泰岳软件股份有限公司 | Return information generates training method, return information generation method and the device of model |
| CN108153913B (en) * | 2018-01-24 | 2020-08-07 | 鼎富智能科技有限公司 | Training method of reply information generation model, reply information generation method and device |
| CN108415893A (en) * | 2018-03-15 | 2018-08-17 | 平安科技（深圳）有限公司 | Poem automatic generation method, device, computer equipment and storage medium |
| CN110309275B (en) * | 2018-03-15 | 2024-06-14 | 北京京东尚科信息技术有限公司 | Dialog generation method and device |
| WO2019174186A1 (en) * | 2018-03-15 | 2019-09-19 | 平安科技（深圳）有限公司 | Automatic poem generation method and apparatus, and computer device and storage medium |
| CN108415893B (en) * | 2018-03-15 | 2019-09-20 | 平安科技（深圳）有限公司 | Poem automatic generation method, device, computer equipment and storage medium |
| CN110309275A (en) * | 2018-03-15 | 2019-10-08 | 北京京东尚科信息技术有限公司 | A kind of method and apparatus that dialogue generates |
| CN110555197A (en) * | 2018-05-31 | 2019-12-10 | 清华大学 | Automatic Chinese poem generating method and device |
| CN110555197B (en) * | 2018-05-31 | 2021-04-02 | 清华大学 | Automatic Chinese poem generating method and device |
| CN109408630B (en) * | 2018-10-17 | 2021-10-29 | 杭州世平信息科技有限公司 | Method for automatically generating court opinions according to description of crime facts |
| CN109408630A (en) * | 2018-10-17 | 2019-03-01 | 杭州世平信息科技有限公司 | A method of law court's viewpoint is automatically generated according to the description of crime fact |
| CN109582956A (en) * | 2018-11-15 | 2019-04-05 | 中国人民解放军国防科技大学 | text representation method and device applied to sentence embedding |
| CN109740145A (en) * | 2018-11-30 | 2019-05-10 | 平安科技（深圳）有限公司 | Lyrics intelligent generation method, device, storage medium and computer equipment |
| CN109815493A (en) * | 2019-01-09 | 2019-05-28 | 厦门大学 | A kind of modeling method that the intelligence hip-hop music lyrics generate |
| CN111767694B (en) * | 2019-03-26 | 2024-04-16 | 北京京东尚科信息技术有限公司 | Text generation method, apparatus and computer readable storage medium |
| CN111767694A (en) * | 2019-03-26 | 2020-10-13 | 北京京东尚科信息技术有限公司 | Text generation method and device and computer readable storage medium |
| CN110164412A (en) * | 2019-04-26 | 2019-08-23 | 吉林大学珠海学院 | A kind of music automatic synthesis method and system based on LSTM |
| CN110362696A (en) * | 2019-06-11 | 2019-10-22 | 平安科技（深圳）有限公司 | Lyrics generation method, system, computer equipment and computer readable storage medium |
| CN112084310A (en) * | 2019-06-12 | 2020-12-15 | 阿里巴巴集团控股有限公司 | Reply information generation and automatic reply method and device |
| CN110287489A (en) * | 2019-06-24 | 2019-09-27 | 北京大米科技有限公司 | Document creation method, device, storage medium and electronic equipment |
| CN110399522A (en) * | 2019-07-03 | 2019-11-01 | 中国传媒大学 | A kind of music singing search method and device based on LSTM and layering and matching |
| US11170176B2 (en) | 2019-09-27 | 2021-11-09 | International Business Machines Corporation | Artificial intelligence based word generation |
| CN110688857A (en) * | 2019-10-08 | 2020-01-14 | 北京金山数字娱乐科技有限公司 | Article generation method and device |
| CN112783334A (en) * | 2019-11-08 | 2021-05-11 | 阿里巴巴集团控股有限公司 | Text generation method and device, electronic equipment and computer-readable storage medium |
| CN112783334B (en) * | 2019-11-08 | 2024-06-11 | 阿里巴巴集团控股有限公司 | Text generation method and device, electronic equipment and computer readable storage medium |
| CN111191439A (en) * | 2019-12-16 | 2020-05-22 | 浙江大搜车软件技术有限公司 | Natural sentence generation method and device, computer equipment and storage medium |
| CN111506767A (en) * | 2020-03-04 | 2020-08-07 | 北京灵动音科技有限公司 | Song word filling processing method and device, electronic equipment and storage medium |
| CN111797196A (en) * | 2020-06-01 | 2020-10-20 | 武汉大学 | Service discovery method combining attention mechanism LSTM and neural topic model |
| CN112163414A (en) * | 2020-08-27 | 2021-01-01 | 同济大学 | Chinese lyric generating method based on Word2Vec, LSTM and attention mechanism |
| CN112163414B (en) * | 2020-08-27 | 2022-09-20 | 同济大学 | Chinese lyric generating method based on Word2Vec, LSTM and attention mechanism |
| CN112669815A (en) * | 2020-12-17 | 2021-04-16 | 广州华多网络科技有限公司 | Song customization generation method and corresponding device, equipment and medium |
| CN112669815B (en) * | 2020-12-17 | 2023-05-19 | 广州华多网络科技有限公司 | Song customization generation method and corresponding device, equipment and medium thereof |
| KR102368418B1 (en) * | 2020-12-31 | 2022-03-02 | (주)휴에버그린팜 | Apparatus and Method for Creating Rap Lyrics included Punch Lines |
| CN115114474A (en) * | 2021-03-17 | 2022-09-27 | 北京达佳互联信息技术有限公司 | Lyric generation model training method, lyric generation method, device and storage medium |
| CN113553863B (en) * | 2021-06-30 | 2023-10-20 | 北京百度网讯科技有限公司 | Text generation method, device, electronic equipment and storage medium |
| CN113553863A (en) * | 2021-06-30 | 2021-10-26 | 北京百度网讯科技有限公司 | Text generation method and device, electronic equipment and storage medium |
| US12106062B2 (en) | 2021-06-30 | 2024-10-01 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method and apparatus for generating a text, and storage medium |
| CN114186545A (en) * | 2021-11-10 | 2022-03-15 | 北京达佳互联信息技术有限公司 | Lyric generation method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106528858A (en) | Lyrics generating method and device | |
| CN105955964B (en) | A kind of method and apparatus automatically generating poem | |
| CN110222164B (en) | Question-answer model training method, question and sentence processing device and storage medium | |
| CN107844469B (en) | Text simplification method based on word vector query model | |
| CN109857845B (en) | Model training and data retrieval method, device, terminal and computer-readable storage medium | |
| US10268671B2 (en) | Generating parse trees of text segments using neural networks | |
| CN109346064B (en) | Training method and system for end-to-end speech recognition model | |
| CN107291836B (en) | Chinese text abstract obtaining method based on semantic relevancy model | |
| KR102139387B1 (en) | Method and apparatus for speech synthesis based on large corpus | |
| CN109657041A (en) | The problem of based on deep learning automatic generation method | |
| CN110705284B (en) | Poetry generation quality optimization method and system based on neural network generation model | |
| JP2019510290A5 (en) | ||
| CN112541060B (en) | End-to-end task type dialogue learning framework and method based on confrontation training | |
| CN110083710A (en) | It is a kind of that generation method is defined based on Recognition with Recurrent Neural Network and the word of latent variable structure | |
| CN108153864A (en) | Method based on neural network generation text snippet | |
| CN108563622B (en) | Absolute sentence generation method and device with style diversity | |
| CN111767694B (en) | Text generation method, apparatus and computer readable storage medium | |
| CN111291175B (en) | Method for automatically generating submitted demand abstract based on strategy gradient algorithm | |
| CN113822054A (en) | Chinese grammar error correction method and device based on data enhancement | |
| CN114860915A (en) | Model prompt learning method and device, electronic equipment and storage medium | |
| CN114048301B (en) | Satisfaction-based user simulation method and system | |
| CN111144553A (en) | Image description method based on space-time memory attention | |
| CN106843523A (en) | Character input method and device based on artificial intelligence | |
| CN113033182A (en) | Text creation auxiliary method and device and server | |
| CN117453898B (en) | Cross-modal question-answering processing method and device based on thinking chain |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170322 |
|
| RJ01 | Rejection of invention patent application after publication |