CN108447475A - A kind of method for building up of the speech recognition modeling based on electric power dispatching system - Google Patents
A kind of method for building up of the speech recognition modeling based on electric power dispatching system Download PDFInfo
- Publication number
- CN108447475A CN108447475A CN201810173642.XA CN201810173642A CN108447475A CN 108447475 A CN108447475 A CN 108447475A CN 201810173642 A CN201810173642 A CN 201810173642A CN 108447475 A CN108447475 A CN 108447475A
- Authority
- CN
- China
- Prior art keywords
- model
- electric power
- dispatching system
- speech recognition
- power dispatching
- 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 31
- 238000013528 artificial neural network Methods 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims description 23
- 239000011159 matrix material Substances 0.000 claims description 21
- 239000000463 material Substances 0.000 claims description 9
- 238000004458 analytical method Methods 0.000 claims description 6
- 238000004088 simulation Methods 0.000 claims description 4
- 230000007704 transition Effects 0.000 claims description 4
- 239000004744 fabric Substances 0.000 claims description 2
- 238000012546 transfer Methods 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 abstract description 7
- 238000002474 experimental method Methods 0.000 abstract description 2
- 230000005540 biological transmission Effects 0.000 abstract 1
- 230000005611 electricity Effects 0.000 abstract 1
- 210000002569 neuron Anatomy 0.000 description 3
- 238000013507 mapping Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000005236 sound signal Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/14—Speech classification or search using statistical models, e.g. Hidden Markov Models [HMMs]
- G10L15/142—Hidden Markov Models [HMMs]
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
- G10L15/063—Training
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/14—Speech classification or search using statistical models, e.g. Hidden Markov Models [HMMs]
- G10L15/142—Hidden Markov Models [HMMs]
- G10L15/148—Duration modelling in HMMs, e.g. semi HMM, segmental models or transition probabilities
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/16—Speech classification or search using artificial neural networks
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
- G10L15/063—Training
- G10L2015/0631—Creating reference templates; Clustering
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
- G10L15/063—Training
- G10L2015/0635—Training updating or merging of old and new templates; Mean values; Weighting
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
- G10L15/063—Training
- G10L2015/0638—Interactive procedures
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Probability & Statistics with Applications (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Machine Translation (AREA)
Abstract
The present invention relates to a kind of method for building up of the speech recognition modeling based on electric power dispatching system, technical characteristics be include step 1, the sound model based on electric power dispatching system is established using HMM model and deep neural network；Step 2 establishes the language model based on electric power dispatching system using statistical language model method.The present invention establishes the exclusive industry acoustic model of power grid using the technology of HMM (Hidden Markov) and DNN (deep neural network) and is based on N Gram speech models, effectively increase the accuracy rate and working efficiency of speech recognition, and eliminate error, the needs of power scheduling industry speech recognition can preferably be met, through in Central China Power Grid experiments have shown that, Central China Power Grid hair transmission of electricity Plan rescheduling, mountain fire scene, primary equipment operates (turning state) and failure reports four operative scenario continuous speech recognitions and tests rate of accuracy reached to 85% or more, it is with a wide range of applications.
Description
Technical field
The invention belongs to technical field of voice recognition, especially a kind of speech recognition modeling based on electric power dispatching system
Method for building up.
Background technology
Speech recognition is to convert speech into the technology of text, is a branch of artificial intelligence natural language processing.Mesh
Before, regular speech identification software can not apply to power scheduling industry, mainly by online speech interface mode to voice content
It is identified, is applied to power scheduling daily record speech recognition accuracy less than 60%, it is difficult to meet the needs of power scheduling.How
Artificial intelligent voice identification technology is applied in power scheduling, during power scheduling, identifies the call between dispatcher
Content, and corresponding order is triggered, it allows machine more to substitute the work of people, improves the intelligent level of management and running, carry
High working efficiency eliminates human error, it is ensured that power network safety operation is current problem in the urgent need to address.
Invention content
It is an object of the invention to overcome the deficiencies in the prior art, propose that a kind of reasonable design, speech recognition be accurate, work
Method for building up efficient and that the speech recognition modeling based on electric power dispatching system for thinking mistake can be effectively eliminated.
The present invention solves its technical problem and following technical scheme is taken to realize：
A kind of method for building up of the speech recognition modeling based on electric power dispatching system, includes the following steps：
Step 1 establishes the sound model based on electric power dispatching system using HMM model and deep neural network；
Step 2 establishes the language model based on electric power dispatching system using statistical language model method.
Further, the concrete methods of realizing of the step 1 includes the following steps：
(1) MFCC features are obtained by recording materials, obtain the matrix of one 13 dimension * frame number；
(2) mixed Gauss model：With the matrix of the 13 dimension * frame numbers of Gauss model area simulation of multidimensional, if the matrix obeys height
This distribution, then looks for mean value and variance matrix；
(3) sound model training is carried out.
Further, the method that (3) step carries out sound model training includes the following steps：
1. segmenting recording text to obtain basic unit, one is indicated using 3-5 above-mentioned basic units in HMM model
Phoneme；
2. initializing HMM model：It is respectively 0 and 1 that the mean value of each phoneme and variance, which is arranged, and transition probability matrix is set as
Both ends are small broad in the middle, and for the HMM model of 3-5 state, each phoneme is divided into 3-5 state；
3. to the HMM model analysis and arrangement after initial, the HMM model of each phoneme is generated；
4. training the HMM model of phoneme level according to training data, following three parameter is obtained：Initial state probabilities point
Cloth, hidden state sequence transfer matrix and hidden state under export observed value probability distribution.
Further, the language model of the step 2 includes dictionary knowledge, the knowledge of grammar and syntactic knowledge three-decker,
It trains to obtain by carrying out grammer, semantic analysis to training text database by statistical language model method.
The advantages and positive effects of the present invention are：
The present invention establishes the exclusive industry sound of power grid using the technology of HMM (Hidden Markov) and DNN (deep neural network)
It learns model and is based on N-Gram speech models, effectively increase the accuracy rate and working efficiency of speech recognition, and eliminate mistake
Difference can preferably meet the needs of power scheduling industry speech recognition, experiments have shown that, Central China Power Grid is sent out defeated through in Central China Power Grid
It is real that electric Plan rescheduling, mountain fire scene, primary equipment operation (turning state) and failure report four operative scenario continuous speech recognitions
Rate of accuracy reached is tested to 85% or more, is with a wide range of applications.
Figure of description
Fig. 1 is that the DNN computing modules of the present invention carry out operation schematic diagram by GPU.
Specific implementation mode
The present invention will be further described with reference to embodiments.
A kind of method for building up of the speech recognition modeling based on electric power dispatching system, including sound model and speech model
It establishes.The present invention uses power grid industry language material DNN model extraction Bottleneck features to replace the short-term spectrum in the model
Feature calculates sufficient statistic, can be than carrying out speech sound signal terminal point detection, using HMM (Hidden Markov) and DNN using frequency
The acoustic model of the exclusive industry of technology building power grid of (deep neural network) and be based on N-Gram (N meta-models) voice mould
Type, to meet the speech recognition accuracy of power scheduling speech recognition system.
The present invention includes the following steps：
Step 1 is established using HMM (Hidden Markov) and DNN (deep neural network) technology based on electric power dispatching system
Sound model.The specific method is as follows：
(1) MFCC features are obtained by recording materials, obtain the matrix of one 13 dimension * frame number.
(2) mixed Gauss model：With the matrix of the 13 dimension * frame numbers of Gauss model area simulation of multidimensional, it is assumed that the matrix is obeyed
Gaussian Profile then looks for mean value and variance matrix.In simulation process, it can be simulated with a multidimensional Gaussian function.
(3) sound model training is carried out：
1. it segments recording text to obtain basic unit, it is above-mentioned basic using 3-5 in HMM (Hidden Markov) model
Unit indicates a phoneme.Simply understand and is exactly：The mean value and variance matrix of each phoneme are it is known that the recording material for passing through us
Material, so that it may to know the transition probability matrix between each phoneme.
2. initializing HMM model：It is respectively 0 and 1 that the mean value of each phoneme and variance, which is arranged, and transition probability matrix is set as
Both ends are small broad in the middle, and for the hmm of 5 states, i.e., each phoneme is divided into 5 states, that is, completes the initialization of HMM for this
Whole process.
3. to the HMM model analysis and arrangement after initial, the HMM model of each phoneme is generated.
4. training the HMM model of phoneme level according to training data：Three parameters will be obtained by training：Original state is general
Rate is distributed π, the shift-matrix A of hidden state sequence (is exactly that some state is transferred in another shape probability of state observation sequence
This mean value or variance probability) and some hidden state under export observed value probability distribution B.
By taking Central China Power Grid as an example, the acoustic model for Central China Power Grid has been built on linux.In the training process, it adopts
Acoustic model is trained with ready recording materials and phoneme.Conventional method establishes acoustic model using GMM+HMM, and the present invention adopts
Acoustic model (as shown in Figure 1) is established with HMM+DNN methods.The present invention uses GPU in acoustic training model so that instruction
Experienced speed is greatly improved, and the accuracy rate of acoustic training model is from 87% to 94%.Its calculation process is as follows：It is first
Audio file is first converted to discrete numerical characteristic MFCC and Fbanks feature, matrix operation is then carried out by DNN
And Nonlinear Mapping, mapping result is subjected to final speech recognition by HMM model.DNN computing modules therein pass through GPU
Carry out operation.According to the voice training language material collected at present and training cross validation results, best super of modelling effect is filtered out
Ginseng.
DNN models can determine the emission probability of HMM.The number of plies of this DNN model will not be less than 5 layers under normal conditions,
Every layer of general thousands of neurons composition.During identifying voice, a bit of voice can all be extracted into Fig. 1 corresponding
Observation, and according to the state computation emission probability (namely comparing similarity with different pronunciations) in HMM, choosing
Emission probability maximum path is selected as final result.There is DNN networks the huge parameter of data to need to learn, each layer network
There is millions of parameters, and the input of next layer network is the output of last layer network, is trained in one under normal conditions
Deng acoustic model need nearly 2,000 CPU cores to run nearly one month.Further, since the DNN used in acoustic model is more
It is special：Each layer of a neuron is dependent on all neurons of last layer, therefore, if model different levels are sliced into
If being trained respectively on different servers, huge network overhead can be brought, keeps system actually unavailable, therefore I
Used GPU during training DNN, and by constantly optimizing so that training speed has closely compared to single server
2000 times of speed improves, to make the training of DNN models become a reality.In order to make DNN models can be applied to the clothes on line
In business, calculating of the DNN on CPU has also been made in the case that optimization lies in and do not influence accuracy rate in we, and calculating speed is promoted
Nearly 10 times.Meanwhile by the application of DNN models, we have dropped 35% or so at the character error rate of speech recognition system.
Step 2 establishes the language based on electric power dispatching system using statistical language model method N-Gram models (N meta-models)
Say model.
The language model of the present invention includes dictionary knowledge, the knowledge of grammar and syntactic knowledge three-decker, to training text
Database carries out grammer, semantic analysis, trains to obtain by being based on statistical model.The language model is for calculating a sentence
The probabilistic model of probability of occurrence is mainly used for determining the possibility bigger of which word sequence, or several words is occurring
In the case of predict the content of next word that will occur；It says it in another way, language model is for constraining word search
, which define which words can follow behind a upper identified word (matching is the processing procedure of a sequence), this
Sample can be to exclude some impossible words for matching process.In addition, it can be effectively combined Chinese grammar and semanteme
Knowledge, the internal relation between descriptor reduce search range to improve discrimination.
Statistical language model be with the method for probability statistics come disclose in linguistic unit statistical law, wherein N-Gram
Model (N meta-models) is simple and effective, is widely used, N meta-models contain the statistics of word sequence.N-Gram models are based on this
Sample is a kind of it is assumed that the appearance of n-th of word is only related to the word of front N-1, and, the probability of whole sentence all uncorrelated to other any words
It is exactly the product of each word probability of occurrence.These probability can be by directly counting N number of word while the number occurred from language material
It obtains.Common N meta-models are the Bi-Gram of binary and the Tri-Gram of ternary.By using to N-Gram algorithms so that I
The accuracy rate of language model improve 33%.
It is emphasized that embodiment of the present invention is illustrative, without being restrictive, therefore packet of the present invention
Include the embodiment being not limited to described in specific implementation mode, it is every by those skilled in the art according to the technique and scheme of the present invention
The other embodiment obtained, also belongs to the scope of protection of the invention.
Claims (4)
1. a kind of method for building up of the speech recognition modeling based on electric power dispatching system, it is characterised in that include the following steps：
Step 1 establishes the sound model based on electric power dispatching system using HMM model and deep neural network；
Step 2 establishes the language model based on electric power dispatching system using statistical language model method.
2. a kind of method for building up of speech recognition modeling based on electric power dispatching system according to claim 1, feature
It is：The concrete methods of realizing of the step 1 includes the following steps：
(1) MFCC features are obtained by recording materials, obtain the matrix of one 13 dimension * frame number；
(2) mixed Gauss model：With the matrix of the 13 dimension * frame numbers of Gauss model area simulation of multidimensional, if the matrix obeys Gauss point
Cloth then looks for mean value and variance matrix；
(3) sound model training is carried out.
3. a kind of method for building up of speech recognition modeling based on electric power dispatching system according to claim 2, feature
It is：(3) method that the step carries out sound model training includes the following steps：
1. segmenting recording text to obtain basic unit, a sound is indicated using 3-5 above-mentioned basic units in HMM model
Element；
2. initializing HMM model：It is respectively 0 and 1 that the mean value of each phoneme and variance, which is arranged, and transition probability matrix is set as both ends
Small broad in the middle, for the HMM model of 3-5 state, each phoneme is divided into 3-5 state；
3. to the HMM model analysis and arrangement after initial, the HMM model of each phoneme is generated；
4. training the HMM model of phoneme level according to training data, following three parameter is obtained：Initial state probabilities are distributed, are hidden
The probability distribution of observed value is exported under transfer matrix and hidden state containing status switch.
4. a kind of method for building up of speech recognition modeling based on electric power dispatching system according to claim 1, feature
It is：The language model of the step 2 includes dictionary knowledge, the knowledge of grammar and syntactic knowledge three-decker, by training
Text database carries out grammer, semantic analysis, and trains to obtain by statistical language model method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810173642.XA CN108447475A (en) | 2018-03-02 | 2018-03-02 | A kind of method for building up of the speech recognition modeling based on electric power dispatching system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810173642.XA CN108447475A (en) | 2018-03-02 | 2018-03-02 | A kind of method for building up of the speech recognition modeling based on electric power dispatching system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108447475A true CN108447475A (en) | 2018-08-24 |
Family
ID=63192885
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810173642.XA Pending CN108447475A (en) | 2018-03-02 | 2018-03-02 | A kind of method for building up of the speech recognition modeling based on electric power dispatching system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108447475A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110689880A (en) * | 2019-10-21 | 2020-01-14 | 国家电网公司华中分部 | Voice recognition method and device applied to power dispatching field |
| CN112420042A (en) * | 2020-11-19 | 2021-02-26 | 国网北京市电力公司 | Control method and device of power system |
| CN112530440A (en) * | 2021-02-08 | 2021-03-19 | 浙江浙达能源科技有限公司 | Intelligent voice recognition system for power distribution network scheduling tasks based on end-to-end model |
| CN112530434A (en) * | 2020-12-21 | 2021-03-19 | 云南电网有限责任公司玉溪供电局 | Automatic intelligent robot on duty scheduling system of power station |
| CN113920990A (en) * | 2021-12-14 | 2022-01-11 | 国网山东省电力公司乳山市供电公司 | Intelligent voice recognition processing system and method for power supply client |
| CN115312038A (en) * | 2022-07-15 | 2022-11-08 | 中电万维信息技术有限责任公司 | Intelligent voice recognition system and method based on communication scheduling instruction |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2011209980A (en) * | 2010-03-30 | 2011-10-20 | Toshiba Corp | Power system monitoring controller |
| CN102867276A (en) * | 2012-08-14 | 2013-01-09 | 俞琳 | Interactive control method based on electric power operational system |
| CN103578464A (en) * | 2013-10-18 | 2014-02-12 | 威盛电子股份有限公司 | Language model establishing method, speech recognition method and electronic device |
| KR20140096844A (en) * | 2013-01-29 | 2014-08-06 | 엘에스산전 주식회사 | Operation method of energy management system using an isolated language voice recognition |
| US20150379988A1 (en) * | 2014-06-26 | 2015-12-31 | Nvoq Incorporated | System and methods to create and determine when to use a minimal user specific language model |
| US20160240190A1 (en) * | 2015-02-12 | 2016-08-18 | Electronics And Telecommunications Research Institute | Apparatus and method for large vocabulary continuous speech recognition |
| US20160253989A1 (en) * | 2015-02-27 | 2016-09-01 | Microsoft Technology Licensing, Llc | Speech recognition error diagnosis |
| CN107301234A (en) * | 2017-06-27 | 2017-10-27 | 国网浙江省电力公司宁波供电公司 | A kind of voice and word interactive system based on regulation and control atypia data |
| CN107680582A (en) * | 2017-07-28 | 2018-02-09 | 平安科技（深圳）有限公司 | Acoustic training model method, audio recognition method, device, equipment and medium |
-
2018
- 2018-03-02 CN CN201810173642.XA patent/CN108447475A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2011209980A (en) * | 2010-03-30 | 2011-10-20 | Toshiba Corp | Power system monitoring controller |
| CN102867276A (en) * | 2012-08-14 | 2013-01-09 | 俞琳 | Interactive control method based on electric power operational system |
| KR20140096844A (en) * | 2013-01-29 | 2014-08-06 | 엘에스산전 주식회사 | Operation method of energy management system using an isolated language voice recognition |
| CN103578464A (en) * | 2013-10-18 | 2014-02-12 | 威盛电子股份有限公司 | Language model establishing method, speech recognition method and electronic device |
| US20150379988A1 (en) * | 2014-06-26 | 2015-12-31 | Nvoq Incorporated | System and methods to create and determine when to use a minimal user specific language model |
| US20160240190A1 (en) * | 2015-02-12 | 2016-08-18 | Electronics And Telecommunications Research Institute | Apparatus and method for large vocabulary continuous speech recognition |
| US20160253989A1 (en) * | 2015-02-27 | 2016-09-01 | Microsoft Technology Licensing, Llc | Speech recognition error diagnosis |
| CN107301234A (en) * | 2017-06-27 | 2017-10-27 | 国网浙江省电力公司宁波供电公司 | A kind of voice and word interactive system based on regulation and control atypia data |
| CN107680582A (en) * | 2017-07-28 | 2018-02-09 | 平安科技（深圳）有限公司 | Acoustic training model method, audio recognition method, device, equipment and medium |
Non-Patent Citations (1)
| Title |
|---|
| 梁静: "基于深度学习的语音识别研究", 《中国优秀硕士学位论文全文数据库-信息科技辑》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110689880A (en) * | 2019-10-21 | 2020-01-14 | 国家电网公司华中分部 | Voice recognition method and device applied to power dispatching field |
| CN112420042A (en) * | 2020-11-19 | 2021-02-26 | 国网北京市电力公司 | Control method and device of power system |
| CN112530434A (en) * | 2020-12-21 | 2021-03-19 | 云南电网有限责任公司玉溪供电局 | Automatic intelligent robot on duty scheduling system of power station |
| CN112530440A (en) * | 2021-02-08 | 2021-03-19 | 浙江浙达能源科技有限公司 | Intelligent voice recognition system for power distribution network scheduling tasks based on end-to-end model |
| CN112530440B (en) * | 2021-02-08 | 2021-05-07 | 浙江浙达能源科技有限公司 | Intelligent voice recognition system for power distribution network scheduling tasks based on end-to-end model |
| CN113920990A (en) * | 2021-12-14 | 2022-01-11 | 国网山东省电力公司乳山市供电公司 | Intelligent voice recognition processing system and method for power supply client |
| CN115312038A (en) * | 2022-07-15 | 2022-11-08 | 中电万维信息技术有限责任公司 | Intelligent voice recognition system and method based on communication scheduling instruction |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108447475A (en) | A kind of method for building up of the speech recognition modeling based on electric power dispatching system | |
| CN110134946B (en) | Machine reading understanding method for complex data | |
| CN101447185B (en) | Audio frequency rapid classification method based on content | |
| CN110379416A (en) | A kind of neural network language model training method, device, equipment and storage medium | |
| CN111475655B (en) | Power distribution network knowledge graph-based power scheduling text entity linking method | |
| CN111353029B (en) | Semantic matching-based multi-turn spoken language understanding method | |
| CN112735383A (en) | Voice signal processing method, device, equipment and storage medium | |
| CN112818105A (en) | Multi-turn dialogue method and system fusing context information | |
| CN113066499B (en) | Method and device for identifying identity of land-air conversation speaker | |
| CN111653275B (en) | Method and device for constructing voice recognition model based on LSTM-CTC tail convolution and voice recognition method | |
| CN111597328B (en) | New event theme extraction method | |
| CN112838946A (en) | Method for constructing intelligent sensing and early warning model based on communication network faults | |
| CN110992959A (en) | Voice recognition method and system | |
| CN112331207B (en) | Service content monitoring method, device, electronic equipment and storage medium | |
| WO2023137918A1 (en) | Text data analysis method and apparatus, model training method, and computer device | |
| CN110704616A (en) | Equipment alarm work order identification method and device | |
| CN113420548A (en) | Entity extraction sampling method based on knowledge distillation and PU learning | |
| CN116110405A (en) | Land-air conversation speaker identification method and equipment based on semi-supervised learning | |
| CN112395891A (en) | Chinese-Mongolian translation method combining Bert language model and fine-grained compression | |
| CN115687934A (en) | Intention recognition method and device, computer equipment and storage medium | |
| CN117172413A (en) | Power grid equipment operation state monitoring method based on multi-mode data joint characterization and dynamic weight learning | |
| CN112882899B (en) | Log abnormality detection method and device | |
| CN113674846A (en) | Hospital intelligent service public opinion monitoring platform based on LSTM network | |
| CN113065352B (en) | Method for identifying operation content of power grid dispatching work text | |
| US20240046921A1 (en) | Method, apparatus, electronic device, and medium for speech processing |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180824 |
|
| WD01 | Invention patent application deemed withdrawn after publication |