CN110349596A - A kind of piano test for confirmation of the grade assessment method and device based on deep learning - Google Patents
A kind of piano test for confirmation of the grade assessment method and device based on deep learning Download PDFInfo
- Publication number
- CN110349596A CN110349596A CN201910529818.5A CN201910529818A CN110349596A CN 110349596 A CN110349596 A CN 110349596A CN 201910529818 A CN201910529818 A CN 201910529818A CN 110349596 A CN110349596 A CN 110349596A
- Authority
- CN
- China
- Prior art keywords
- piano
- audio data
- confirmation
- deep learning
- neural network
- 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.)
- Withdrawn
Links
- 238000012790 confirmation Methods 0.000 title claims abstract description 36
- 238000012360 testing method Methods 0.000 title claims abstract description 36
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000013135 deep learning Methods 0.000 title claims abstract description 23
- 238000011156 evaluation Methods 0.000 claims abstract description 60
- 239000011159 matrix material Substances 0.000 claims abstract description 28
- 238000003062 neural network model Methods 0.000 claims abstract description 26
- 238000012549 training Methods 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 20
- 238000004422 calculation algorithm Methods 0.000 claims description 17
- 238000013528 artificial neural network Methods 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 10
- 238000005457 optimization Methods 0.000 claims description 7
- 230000000306 recurrent effect Effects 0.000 claims description 6
- 238000004891 communication Methods 0.000 claims description 3
- 235000013399 edible fruits Nutrition 0.000 claims 1
- 238000000605 extraction Methods 0.000 claims 1
- 229910000831 Steel Inorganic materials 0.000 description 3
- 239000010959 steel Substances 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 241000282994 Cervidae Species 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000003796 beauty Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000001351 cycling effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005553 drilling Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 229940061368 sonata Drugs 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000003786 synthesis reaction Methods 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
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
-
- 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
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
- G10L25/24—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters the extracted parameters being the cepstrum
-
- 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
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
- G10L25/30—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique using 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
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
- G10L25/51—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Auxiliary Devices For Music (AREA)
Abstract
The invention discloses a kind of piano test for confirmation of the grade assessment method and device based on deep learning, includes the following steps: the audio data for obtaining piano melody to be evaluated；The audio data is pre-processed；Acoustic feature is extracted from pretreated audio data；The acoustic feature is inputted into trained ResCNN neural network model in advance, obtains prediction probability matrix P；The evaluation result of examining for the levels of the piano melody to be evaluated is determined according to the prediction probability matrix P.The present invention provides a kind of piano test for confirmation of the grade assessment method and device based on deep learning, can simulate piano test for confirmation of the grade profession scoring teacher with entered oneself for the examination corresponding to piano melody to be evaluated the evaluation criteria of grade treat evaluation piano melody evaluate, and then determine the evaluation result of examining for the levels of piano melody to be evaluated, substantially increase the efficiency that scores of examining for the levels, human resources are saved, while can effectively avoid influence of the artificial subjective factor to evaluation result of examining for the levels.
Description
Technical field
The present invention relates to a kind of piano test for confirmation of the grade assessment method and device based on deep learning.
Background technique
Piano is a kind of keyboard instrument from western classicality musical instrument, is made of 88 keys and metallic cord soundboard, bullet
The person of playing then taps steel wire string and makes a sound by affecting the gavel inside piano by the key on lower keyboard.Piano is because of it
Wide range, the absolutely tone color of beauty, are known as the king of musical instrument.As the improvement of people's living standards, the people of study piano performance
It is more and more, and the amateurish piano study person for participating in piano test for confirmation of the grade is also more and more.
With the continuous rising of student's quantity of piano test for confirmation of the grade, the quantity of piano test for confirmation of the grade profession scoring teacher also increases therewith
It is more, while higher requirements are also raised to profession scoring teacher, the operating pressure of profession scoring teacher can also increase with it.It removes
Except this, student carries out the scoring inefficiency of piano test for confirmation of the grade, and appraisal result of examining for the levels is easy personal subjective by profession scoring teacher
The influence of factor leads to that an objective appraisal cannot be provided according to the case where performance.Obviously, commenting for piano test for confirmation of the grade how is improved
Component efficiency, the investment for reducing piano test for confirmation of the grade human resources become piano music and examine for the levels in development process one and urgently to be resolved ask
Topic.
Summary of the invention
The present invention provides a kind of piano test for confirmation of the grade assessment method and device based on deep learning, which overcome the prior arts
Piano test for confirmation of the grade present in shortcoming.
The technical solution adopted by the present invention to solve the technical problems is: a kind of piano test for confirmation of the grade evaluation based on deep learning
Method includes the following steps: for judging whether piano melody to be evaluated meets the corresponding evaluation criteria for entering oneself for the examination grade
Obtain the audio data of piano melody to be evaluated；
The audio data is pre-processed；
Acoustic feature is extracted from pretreated audio data；
The acoustic feature is inputted into trained ResCNN neural network model in advance, obtains prediction probability matrix P；
The evaluation result of examining for the levels of the piano melody to be evaluated is determined according to the prediction probability matrix P.
Further, described to extract acoustic feature from pretreated audio data specifically: to be fallen using mel-frequency
Spectral coefficient extracts acoustic feature from pretreated audio data；
The acoustic feature is by MFCC feature, single order MFCC feature, second order MFCC feature, energy feature, single order energy
The multidimensional assemblage characteristic that feature and secondary energy feature are combined into.
Further, described that the audio data pre-process including normalized, mute section of processing of removal, is gone
It makes an uproar any one or more in handling.
Further, mute section of processing of the removal specifically: each frame of the audio data is calculated using vad algorithm
Energy value, frame of the energy value lower than energy threshold TH is mute frame, is otherwise speech frame, gives up the mute frame, described in reservation
Speech frame.
Further, the denoising is using the audio defeat algorithm based on Recognition with Recurrent Neural Network RNN to the audio
Data are denoised.
Further, the evaluation knot of examining for the levels that the piano melody to be evaluated is determined according to the prediction probability matrix P
Fruit specifically: the evaluation result of examining for the levels includes failing, passing, is good and outstanding, each of described prediction probability matrix P
The value of element represents the probability that the piano melody to be evaluated belongs to the corresponding evaluation result of examining for the levels of the element, and the prediction is general
Corresponding final examine for the levels evaluation of the evaluation result as the piano melody to be evaluated of examining for the levels of the maximum element of rate matrix P intermediate value
As a result.
Further, the training step of the trained ResCNN neural network model in advance includes:
Training sample set is obtained, the training sample set includes sample audio data known to multiple evaluation results；
The sample audio data is pre-processed；
Sample acoustic feature is extracted from pretreated sample audio data；
Construct ResCNN neural network model；
Using sample acoustic feature training ResCNN neural network model, trained ResCNN neural network is obtained
Model.
Further, described using sample acoustic feature training ResCNN neural network model, it obtains trained
ResCNN neural network model, includes the following steps:
The sample acoustic feature is inputted in ResCNN neural network model, obtains prediction probability matrix P ', and according to
Prediction probability matrix P ' determines the prediction evaluation result of the sample audio data；
The prediction evaluation result of the sample audio data is compared with its true evaluation result, utilizes loss function
Calculate penalty values；
Model parameter is updated using loss function optimization algorithm according to the penalty values；
It repeats the above steps and is iterated training, until loss function value convergence or the number of iterations reach predetermined time
Number completes the training of ResCNN neural network model.
Further, the loss function uses cross entropy loss function, and the loss function optimization algorithm is using random
Gradient descent algorithm.
A kind of piano test for confirmation of the grade evaluation device, comprising:
Recording device, for obtaining the audio data of piano melody to be evaluated and being sent to computer equipment；
Computer equipment, the computer equipment include at least one processor and at least one described processor communication
The memory of connection；Wherein, the memory is stored with the instruction that can be executed by least one described processor, described instruction quilt
At least one described processor executes, so that at least one described processor executes a kind of above-mentioned piano based on deep learning
It examines for the levels assessment method.
Compared to the prior art, the invention has the following advantages:
1, the present invention provides a kind of piano test for confirmation of the grade assessment method and device based on deep learning, can simulate piano and examine
Grade profession scoring teacher treats evaluation piano melody progress to enter oneself for the examination the evaluation criteria of grade corresponding to piano melody to be evaluated
Evaluation, and then determine the evaluation result of examining for the levels of piano melody to be evaluated, the efficiency that scores of examining for the levels is substantially increased, manpower money is saved
Source reduces human cost, while can effectively avoid influence of the artificial subjective factor to evaluation result of examining for the levels.
2, the present invention uses input of the multidimensional assemblage characteristic as ResCNN neural network model, can have different characteristic
Effect combines, more can accurate judgement piano melody evaluate whether meet the corresponding evaluation criteria for entering oneself for the examination grade.
Invention is further described in detail with reference to the accompanying drawings and embodiments；But one kind of the invention is based on depth
The piano test for confirmation of the grade assessment method and device of habit are not limited to the embodiment.
Detailed description of the invention
Fig. 1 is the processing step flow chart of the method for the present invention.
The reference numerals are as follows:
Specific embodiment
Below with reference to attached drawing of the present invention, technical solution in the embodiment of the present invention is described in detail and discusses.It answers
Work as understanding, described herein specific examples are only used to explain the present invention, is not intended to limit the present invention.
Big problem is put into for traditional piano test for confirmation of the grade assessment method scoring inefficiency, human resources, the present invention mentions
A kind of piano test for confirmation of the grade assessment method based on deep learning has been supplied, teacher is scored with steel to be evaluated by simulation piano test for confirmation of the grade profession
The evaluation criteria that grade is entered oneself for the examination corresponding to qin melody is treated evaluation piano melody and is evaluated, and then determines that piano to be evaluated is happy
Bent evaluation result of examining for the levels.
The method of the invention and device are carried out using ten grades of piano of evaluation of examining for the levels as specific embodiment as follows detailed
Description, certainly, the evaluation of examining for the levels of other piano grades (such as any level-one in one to nine grades of piano) can also be by the present embodiment
Technical solution is implemented.
A kind of piano test for confirmation of the grade assessment method based on deep learning shown in Figure 1, of the invention, including walk as follows
It is rapid:
Step S1: the audio data of piano melody to be evaluated is obtained.
In the present embodiment, the corresponding grade of entering oneself for the examination of the piano melody to be evaluated is ten grades of piano, specifically, described to be evaluated
Determine piano melody be by examine for the levels ten grades of student's playing piano examine for the levels song when recorded to obtain, ten grades of piano examine
Grade song includes but is not limited to big brilliant big waltz (Op.34No.3), the Men Deer for adjusting sonata, Chopin of Scarlatti D
Loose scherzo (Op.16No.4).Certainly, in other embodiments, the corresponding grade of entering oneself for the examination of the piano melody to be evaluated can be
Other piano grades (any level-one in such as one to nine grades of piano), the piano melody to be evaluated are by drilling in the student that examines for the levels
Play accordingly enter oneself for the examination grade examine for the levels song when recorded to obtain.
Step S2: pre-processing the audio data, specifically includes normalized, mute section of processing of removal, goes
It makes an uproar any one or more in handling.
The normalized includes the following steps:
The audio data is read, audio frame number evidence is obtained from the audio data；
The audio frame number evidence that will acquire is converted to Audio Matrix；
Normalize Audio Matrix: by Audio Matrix multiplied by normalization coefficientAudio data after being normalized.
Mute section of processing of the removal specifically: calculated using the VAD (voice activity detector) based on energy
Method calculates the energy value of each frame of the audio data, otherwise it is voice that frame of the energy value lower than energy threshold TH, which is mute frame,
Frame gives up the mute frame, retains the speech frame.The value range of the energy threshold TH is 0.003db-0.008db.
The denoising removes the audio data using the audio defeat algorithm based on Recognition with Recurrent Neural Network RNN
It makes an uproar, the Recognition with Recurrent Neural Network RNN (Recurrent Neural Network, RNN) is a kind of with sequence (sequence) number
According to input, recurrence (recursion) is carried out in the evolution tendency of sequence and all nodes (cycling element) are by chain type connection
Recurrent neural network (recursive neural network).
The present invention can be effectively removed mute by being removed mute section of processing or denoising to the audio data
Or the interference of noise, improve the accuracy rate of piano test for confirmation of the grade evaluation result.
Step S3: extracting acoustic feature from pretreated audio data, specifically:
Acoustic feature is extracted from pretreated audio data using mel-frequency cepstrum coefficient (MFCC), that is, uses plum
Pretreated audio data is converted into time-frequency domain information from time-domain signal by that frequency cepstral coefficient；The acoustic feature
For by MFCC feature, single order MFCC feature, second order MFCC feature, energy feature, single order energy feature and secondary energy feature group
The multidimensional assemblage characteristic of synthesis.The present invention uses input of the multidimensional assemblage characteristic as ResCNN neural network model, can incite somebody to action
Different characteristic effectively combines, and effectively improves piano melody and examines for the levels the accuracy rate of evaluation result.
Specifically, after extracting MFCC feature and energy feature in pretreated audio data, it is obtained by calculation one
Rank MFCC feature, second order MFCC feature, single order energy feature and secondary energy feature.In the present embodiment, the acoustic feature is
By 12 dimension MFCC features, 12 dimension single order MFCC features, 12 dimension second order MFCC features, 1 dimension energy feature, 1 dimension single order energy feature
39 dimension assemblage characteristics being combined into 1 dimension secondary energy feature.
Step S4: the acoustic feature is inputted into trained ResCNN neural network model in advance, obtains prediction probability
Matrix P.
Step S5: the evaluation result of examining for the levels of the piano melody to be evaluated is determined according to the prediction probability matrix P, specifically
Are as follows: the evaluation result of examining for the levels includes failing, passing, is good and outstanding, each element in the prediction probability matrix P
Value represents the probability that the piano melody to be evaluated belongs to the corresponding evaluation result of examining for the levels of the element, by the prediction probability matrix
Corresponding final examine for the levels evaluation result of the evaluation result as the piano melody to be evaluated of examining for the levels of the maximum element of P intermediate value.
Specifically, the prediction probability matrix P is expressed as P=[p1,p2,p3,p4], wherein probability value p1、p2、p3And p4Point
Not Wei the piano melody be evaluated belong to fail, pass, good and outstanding probability.
In the present embodiment, the trained ResCNN neural network model in advance is to be instructed for ten grades of examinations of piano
Experienced prediction model, training step are specific as follows:
Training sample set is obtained, the training sample set includes sample audio data known to multiple evaluation results, that is, is instructed
Practice sample set in each sample audio data ten grades of piano examination in evaluation result (evaluation result include fail,
Pass, be good and four kinds outstanding) it is known.
The sample audio data is pre-processed；
Sample acoustic feature is extracted from pretreated sample audio data, specially uses mel-frequency cepstrum coefficient
(MFCC) sample acoustic feature is extracted from pretreated sample audio data；
Construct ResCNN neural network model；
Using sample acoustic feature training ResCNN neural network model, trained ResCNN neural network is obtained
Model；
Wherein, to the sample audio data carry out pretreatment with it is aforementioned treat evaluation piano melody audio data into
Capable pretreatment is identical, also including any one or more in normalized, mute section of processing of removal, denoising；Institute
Sample acoustic feature is stated also by 12 dimension MFCC features, 12 dimension single order MFCC features, 12 dimension second order MFCC features, 1 dimension energy
The 39 dimension assemblage characteristics that measure feature, 1 dimension single order energy feature and 1 dimension secondary energy feature are combined into.
It is described to train ResCNN neural network model using the sample acoustic feature, obtain trained ResCNN nerve
Network model includes the following steps:
The sample acoustic feature is inputted in ResCNN neural network model, obtains prediction probability matrix P ', and according to
Prediction probability matrix P ' determines the prediction evaluation result of the sample audio data, specifically: the prediction probability matrix P ' table
It is shown as P '=[p '1,p′2,p′3,p′4], wherein probability value p '1、p′2、p′3With p '4The respectively described sample audio data belongs to
It fails, pass, good and outstanding probability, by the corresponding evaluation result of the maximum element of the prediction probability matrix P ' intermediate value
(evaluation result includes the prediction evaluation result failing, pass, is good and four kinds outstanding) as the sample audio data.
The prediction evaluation result of the sample audio data is compared with true evaluation result, utilizes loss function meter
Calculate penalty values；
Model parameter is updated using loss function optimization algorithm according to the penalty values；
It repeats the above steps and is iterated training, continuous correction model parameter, until the loss function value is restrained or changed
Generation number reaches pre-determined number, completes the training of ResCNN neural network model.
In the present embodiment, the loss function uses cross entropy loss function, and the loss function optimization algorithm is random
Gradient descent algorithm minimizes cross entropy loss function by using stochastic gradient descent algorithm, improves trained accuracy.When
So, other algorithms can also be used in loss function optimization algorithm, and specific embodiment is not limited thereto.
In the present embodiment, piano melody and different grades of that the training sample set can be played from passing examination personnel
Obtained in piano melody comprising in ten grades of examinations of piano evaluation result be fail, pass, good and outstanding piano it is happy
It is bent.Certainly, in other embodiments, if the corresponding grade of entering oneself for the examination of the piano melody to be evaluated is other piano grade (such as steel
Any level-one in one to nine grades of qin), then the training sample set include in other piano grade examinations evaluation result be
It fails, pass, good and outstanding piano melody.
The present invention also provides a kind of piano test for confirmation of the grade evaluation devices, comprising:
Recording device, for obtaining the audio data of piano melody to be evaluated and being sent to computer equipment, the recording
Device is microphone or microphone array；
Computer equipment, the computer equipment include at least one processor and at least one described processor communication
The memory of connection；Wherein, the memory is stored with the instruction that can be executed by least one described processor, described instruction quilt
At least one described processor executes, so that at least one described processor executes a kind of above-mentioned piano based on deep learning
It examines for the levels assessment method.
Above-described embodiment is only used to further illustrate a kind of piano test for confirmation of the grade assessment method based on deep learning of the invention
And device, but the invention is not limited to embodiment, to the above embodiments according to the technical essence of the invention
What simple modification, equivalent change and modification, falls within the scope of protection of technical solution of the present invention.
Claims (10)
1. a kind of piano test for confirmation of the grade assessment method based on deep learning, which comprises the steps of:
Obtain the audio data of piano melody to be evaluated；
The audio data is pre-processed；
Acoustic feature is extracted from pretreated audio data；
The acoustic feature is inputted into trained ResCNN neural network model in advance, obtains prediction probability matrix P；
The evaluation result of examining for the levels of the piano melody to be evaluated is determined according to the prediction probability matrix P.
2. a kind of piano test for confirmation of the grade assessment method based on deep learning according to claim 1, which is characterized in that it is described from
Acoustic feature is extracted in pretreated audio data specifically: uses mel-frequency cepstrum coefficient from pretreated audio number
According to middle extraction acoustic feature；
The acoustic feature is by MFCC feature, single order MFCC feature, second order MFCC feature, energy feature, single order energy feature
The multidimensional assemblage characteristic being combined into secondary energy feature.
3. a kind of piano test for confirmation of the grade assessment method based on deep learning according to claim 1, which is characterized in that described right
It includes normalized, removal mute section of processing, any one in denoising or more that the audio data, which carries out pretreatment,
?.
4. a kind of piano test for confirmation of the grade assessment method based on deep learning according to claim 3, which is characterized in that described to go
Except mute section of processing specifically: calculate the energy value of each frame of the audio data using vad algorithm, energy value is lower than energy door
The frame for limiting TH is mute frame, is otherwise speech frame, gives up the mute frame, retains the speech frame.
5. a kind of piano test for confirmation of the grade assessment method based on deep learning according to claim 3, which is characterized in that described to go
Processing of making an uproar denoises the audio data using the audio defeat algorithm based on Recognition with Recurrent Neural Network RNN.
6. a kind of piano test for confirmation of the grade assessment method based on deep learning according to claim 1, which is characterized in that described
The evaluation result of examining for the levels of the piano melody to be evaluated is determined according to the prediction probability matrix P specifically: the evaluation knot of examining for the levels
Fruit includes failing, passing, is good and outstanding, and the value of each element in the prediction probability matrix P represents described wait evaluate
Piano melody belongs to the probability of the corresponding evaluation result of examining for the levels of the element, by the maximum element of the prediction probability matrix P intermediate value
Corresponding final examine for the levels evaluation result of the evaluation result as the piano melody to be evaluated of examining for the levels.
7. a kind of piano test for confirmation of the grade assessment method based on deep learning according to claim 1, which is characterized in that described pre-
First the training step of trained ResCNN neural network model includes:
Training sample set is obtained, the training sample set includes sample audio data known to multiple evaluation results；
The sample audio data is pre-processed；
Sample acoustic feature is extracted from pretreated sample audio data；
Construct ResCNN neural network model；
Using sample acoustic feature training ResCNN neural network model, trained ResCNN neural network mould is obtained
Type.
8. a kind of piano test for confirmation of the grade assessment method based on deep learning according to claim 7, which is characterized in that the benefit
With sample acoustic feature training ResCNN neural network model, trained ResCNN neural network model is obtained, including
Following steps:
The sample acoustic feature is inputted in ResCNN neural network model, obtains prediction probability matrix P ', and according to prediction
Probability matrix P ' determines the prediction evaluation result of the sample audio data；
The prediction evaluation result of the sample audio data is compared with its true evaluation result, is calculated using loss function
Penalty values out；
Model parameter is updated using loss function optimization algorithm according to the penalty values；
It repeats the above steps and is iterated training, until loss function value convergence or the number of iterations reach pre-determined number, it is complete
At the training of ResCNN neural network model.
9. a kind of piano test for confirmation of the grade assessment method based on deep learning according to claim 8, which is characterized in that the damage
It loses function and uses cross entropy loss function, the loss function optimization algorithm uses stochastic gradient descent algorithm.
10. a kind of piano test for confirmation of the grade evaluation device characterized by comprising
Recording device, for obtaining the audio data of piano melody to be evaluated and being sent to computer equipment；
Computer equipment, the computer equipment include at least one processor and connect at least one described processor communication
Memory；Wherein, the memory is stored with the instruction that can be executed by least one described processor, and described instruction is described
At least one processor executes, so that at least one described processor perform claim requires a kind of base described in any one of 1-9
In the piano test for confirmation of the grade assessment method of deep learning.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910529818.5A CN110349596A (en) | 2019-06-19 | 2019-06-19 | A kind of piano test for confirmation of the grade assessment method and device based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910529818.5A CN110349596A (en) | 2019-06-19 | 2019-06-19 | A kind of piano test for confirmation of the grade assessment method and device based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110349596A true CN110349596A (en) | 2019-10-18 |
Family
ID=68182362
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910529818.5A Withdrawn CN110349596A (en) | 2019-06-19 | 2019-06-19 | A kind of piano test for confirmation of the grade assessment method and device based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110349596A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117863175A (en) * | 2023-12-25 | 2024-04-12 | 之江实验室 | Offline evaluation system and method for playing piano robot |
-
2019
- 2019-06-19 CN CN201910529818.5A patent/CN110349596A/en not_active Withdrawn
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117863175A (en) * | 2023-12-25 | 2024-04-12 | 之江实验室 | Offline evaluation system and method for playing piano robot |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Gevaert et al. | Neural networks used for speech recognition | |
| CN105719661B (en) | A kind of stringed musical instrument performance sound quality automatic distinguishing method | |
| CN109727608B (en) | Chinese speech-based ill voice evaluation system | |
| CN108520753A (en) | Voice lie detection method based on the two-way length of convolution memory network in short-term | |
| CN110675881B (en) | Voice verification method and device | |
| CN107910020A (en) | Sound of snoring detection method, device, equipment and storage medium | |
| Fan et al. | Singing voice separation and pitch extraction from monaural polyphonic audio music via DNN and adaptive pitch tracking | |
| CN103054586B (en) | Chinese speech automatic audiometric method based on Chinese speech audiometric dynamic word list | |
| Matveev | The problem of voice template aging in speaker recognition systems | |
| CN109346087A (en) | Fight the method for identifying speaker and device of the noise robustness of the bottleneck characteristic of network | |
| CN106531174A (en) | Animal sound recognition method based on wavelet packet decomposition and spectrogram features | |
| CN111986679A (en) | Speaker confirmation method, system and storage medium for responding to complex acoustic environment | |
| Ramashini et al. | Robust cepstral feature for bird sound classification | |
| CN110211604A (en) | A kind of depth residual error network structure for voice deformation detection | |
| Tsenov et al. | Speech recognition using neural networks | |
| CN109584904A (en) | The sightsinging audio roll call for singing education applied to root LeEco identifies modeling method | |
| CN113111151A (en) | Cross-modal depression detection method based on intelligent voice question answering | |
| CN110136746B (en) | Method for identifying mobile phone source in additive noise environment based on fusion features | |
| Usman | On the performance degradation of speaker recognition system due to variation in speech characteristics caused by physiological changes | |
| CN110349596A (en) | A kind of piano test for confirmation of the grade assessment method and device based on deep learning | |
| Marcinek et al. | N-MTTL SI Model: Non-Intrusive Multi-Task Transfer Learning-Based Speech Intelligibility Prediction Model with Scenery Classification. | |
| Sahoo et al. | Detection of speech-based physical load using transfer learning approach | |
| Suwannakhun et al. | Characterizing depressive related speech with MFCC | |
| Deb et al. | Analysis of out-of-breath speech for assessment of person’s physical fitness | |
| Anami et al. | A novel convolutional neural network model for musical instruments’ classification: A deep signal processing approach |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20191018 |