In a preferred embodiment, the system further comprises asuggestion module 5, wherein thesuggestion module 5 is connected to theevaluation module 4, and is configured to provide corresponding suggestions to the singer after providing evaluations according to the singing type matching rate, the sound velocity matching rate, the sound intensity change matching rate, and the pitch correct score.

Specifically, as shown in fig. 1, after the evaluation is made according to the singing type matching rate, the sound velocity matching rate, the sound intensity variation matching rate and the pitch correct score, the singer is suggested accordingly, for example, by taking the 18nd bar 2 in the song "beijing welcome you", the audio of the original song should be from low to high, but the singer may sing from high to low by mistake, so that the singer is suggested to pay attention to the mastering of the sound intensity, and thus the 2 nd bar is repeatedly exercised for the 18 nd bar, so that the singing level of the singer is improved.

step S1, aprocessing module 1 is adopted to process the type of each original song so as to extract a first characteristic parameter of the original song;

step S2, atraining module 2 is adopted for respectively training the sound speed and the sound intensity change of each segmented audio of each original song so as to respectively extract second characteristic parameters of the sound speed and the sound intensity change of each segmented audio;

step S3, analgorithm module 3 is adopted for calculating a final score according to the first characteristic parameter, the second characteristic parameter and a third characteristic parameter of a corresponding time interval of a singing audio;

and step S4, adopting anevaluation module 4 for proposing corresponding evaluation according to the final score.

Through the evaluation method provided above, as shown in fig. 5, firstly, theprocessing module 1 is adopted to classify all the original songs, and the type tempo of each original song can be obtained as, for example, 2/4, 4/4, 3/4, 3/8, 6/8, 9/8, 2/2, 1/4, or the type speed of each original song can be obtained as, for example, 50-140 beats per minute (where each 10 beats is a classification), so that the type or style of each original song can be obtained as: the original song is cheerful or fluent or moderate or relaxed or hypertonic or intense or hot and the like, for example, the type beat of the original song 'love filling the world' is 4/4, the type speed is 76(70-79) beats per minute, so that the type or style of the original song is acquired as relaxed, and the original song is further processed to extract a first characteristic parameter of the original song for subsequent calculation and scoring.

Further, thetraining module 2 is adopted to separately train the sound speed and the sound intensity change of each segmented audio of each original song, in this embodiment, the adopted training methods are CNN (Convolutional Neural networks) and ResNet (Residual Neural networks Residual error Network), so as to extract a large number of second characteristic parameters of the sound speed and the sound intensity change of each segmented audio respectively.

Further, thealgorithm module 3 is adopted to calculate the first characteristic parameter obtained by theprocessing module 1, the second characteristic parameter obtained by thetraining module 2 and the third characteristic parameter of the singing audio in the corresponding time period, so as to obtain the final score of the live singing song.

Furthermore, anevaluation module 4 is adopted to give corresponding evaluation to the singer according to the total score obtained by thealgorithm module 3, so as to help the student improve the singing level.

In a preferred embodiment, step S1 includes:

step S10, a collectingunit 10 is adopted for collecting each original song;

step S11, aclassification unit 11 is adopted for classifying the type of each original song after collection;

step S12, afirst training unit 12 is adopted for training the classified original song;

step S13, a first extractingunit 13 is adopted to extract a first feature parameter according to the training result of thefirst training unit 12;

step S14, a first calculatingunit 14 is adopted to calculate the first characteristic parameter and the third characteristic parameter of the singing audio corresponding time interval, so as to obtain a singing type matching rate.

Specifically, as shown in fig. 6, in step S1, thecollection unit 10 is first used to collect all the original songs, then the types of each collected original song are classified, and the beat and speed of each song are obtained to know which type or style each original song is;

further, thefirst training unit 11 is adopted to train the classified original song by adopting CNN (Convolutional Neural Networks) and ResNet (Residual Neural Networks) algorithms, and then thefirst extraction unit 13 is adopted to extract a large number of first characteristic parameters.

Further, the first calculatingunit 14 is adopted to perform cosine distance calculation on the first characteristic parameter and the third characteristic parameter of the singing audio in the corresponding time period so as to obtain a singing type matching rate and further obtain a style suggestion.

In a preferred embodiment, step S2 includes:

step S20, asegmentation unit 20 is adopted for segmenting each original song according to the beat of the music score to form segmented audio;

step S21, asecond training unit 21 is adopted to train the variation of the sound velocity and the sound intensity in the segmented audio respectively, so as to extract the second characteristic parameters of the variation of the sound velocity and the sound intensity of each segmented audio respectively.

Specifically, as shown in fig. 7, in step S2, thesegmentation unit 20 is first used to segment each original song according to the tempo of the music score to form each small segmented audio, and thesecond training unit 21 is then used to perform separate training on the sound velocity and the sound intensity variation of each small segmented audio respectively to extract a large number of second feature parameters of the sound velocity and the sound intensity variation of each segmented audio respectively.

In a preferred embodiment, step S3 includes:

step S30, using a first comparingunit 30 for comparing the second characteristic parameter of sound velocity and sound intensity variation with the third characteristic parameter of the corresponding time interval of the singing audio to obtain a sound velocity matching rate and a sound intensity variation matching rate;

step S31, using a second calculatingunit 31 for calculating the pitch value of each beat of the singing audio;

step S32, asecond comparison unit 32 is adopted to compare the pitch value of each beat of the singing audio with the pitch value of each beat of the original singing song to obtain a correct pitch score;

step S33, a third calculatingunit 33 is adopted to calculate the final score according to the singing style matching rate, the sound velocity matching rate, the sound intensity variation matching rate and the pitch correct score.

Specifically, as shown in fig. 8, in step S3, thefirst comparison unit 30 is first adopted to compare the second characteristic parameters of the variation of the sound velocity and the sound intensity with the third characteristic parameters of the corresponding time period of the singing audio, so as to obtain the sound velocity matching rate and the sound intensity variation matching rate.

Further, the second calculatingunit 31 calculates a pitch value of each beat of the singing audio by using FFT (fast Fourier transform), the second comparingunit 32 compares the pitch value of each beat of the singing audio with the pitch value of each beat of the original song, if the pitch value of each beat of the singing audio is correct, the score is 1, and if the pitch value of each beat of the singing audio is incorrect, the score is 0, so that a correct pitch score is calculated, and the third calculatingunit 33 obtains the final score according to the following formula:

wherein S represents the final score;

MaxS represents the total score;

o represents a sound velocity matching rate;

p represents a sound intensity change matching rate;

n represents the pitch correct score;

x represents the number of bars of the song;

y represents the number of beats of the song;

j represents the score weight for the pitch correct score.

In a preferred embodiment, the method further comprises a step S5 of using asuggestion module 5 for providing corresponding suggestions to the singer after making evaluations according to the singing type matching rate, the sound velocity matching rate, the sound intensity variation matching rate and the pitch correct score.

Specifically, as shown in fig. 5, after the evaluation is made according to the singing type matching rate, the sound velocity matching rate, the sound intensity variation matching rate and the pitch correct score, thesuggestion module 5 is adopted to correspondingly give suggestions to the singer, for example, by taking the 18 thminor bar 2 in the song "beijing welcome you", the audio of the original sung song should be from low to high, but the singer may possibly sing from high to low, so that the singer can be suggested to pay attention to the mastering of the sound intensity, and thus the 2 nd minor bar 2-time exercise is repeated, so that the singing level of the singer is improved.

While the invention has been described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the spirit and scope of the invention.

Claims

1. A system for evaluating a song performed, comprising:

2. The system of claim 1, wherein the processing module comprises:

a collecting unit for collecting each original song;

3. The system of claim 1, wherein the training module comprises:

4. The system of claim 2, wherein the algorithm module comprises:

5. The system for evaluating a sung song according to claims 2 and 4, further comprising a suggestion module, wherein the suggestion module is connected to the evaluation module, and is configured to suggest to a singer a corresponding suggestion after proposing an evaluation according to the sung type matching rate, the sound velocity matching rate, the sound intensity variation matching rate, and the pitch correct score.

6. A method for evaluating a singing song, comprising the system for evaluating a singing song according to any one of claims 1 to 5, the method comprising:

7. The method of claim 6, wherein the step S1 includes:

step S10, a collecting unit is adopted for collecting each original song;

8. The method of claim 6, wherein the step S2 includes:

9. The method of claim 7, wherein the step S3 includes:

10. The method as claimed in claim 7 or 9, further comprising step S5, wherein a suggestion module is used for providing corresponding suggestions to the singer after the evaluation is made according to the singing genre matching rate, the sound velocity matching rate, the sound intensity variation matching rate and the pitch correct score.